An Alternative Framework to Investigating and Understanding Intraindividual Processes in Substance Abuse Recovery

Abstract

Background:

Sustained recovery from substance abuse is a dynamic intraindividual-level process.

Objectives:

We argue that research on recovery process will benefit from a theoretical approach that captures both the dynamic and the idiographic nature of substance abuse recovery. In addition to setting out why we believe that research on recovery can benefit from such an approach, we provide a demonstration of idiographic within-individual analyses of between- and within-day associations among negative affect, substance use craving, and positive social experiences.

Design and Subjects:

The data used were drawn from 39 abstinent young adults in 12-step recovery from substance abuse (mean age = 22.9, females = 12). Participants provided an average of 26.7 days of daily diary data by end-of-day collections. Unified first-order structural equation models were fit individually to predict daily levels of craving and negative affect from the previous day’s same two variables as well as from both the previous day’s and the same day’s positive social experiences.

Results:

Model estimates demonstrated substantial interindividual heterogeneity in their day-to-day associations in both direction and magnitude, highlighting the importance of applying idiographic approach to understanding recovery. Cluster analyses were subsequently applied to individual model estimates to identify homogeneous subgroups that demonstrated similar day-to-day association patterns, revealing two distinct subgroups that appeared to manage daily abstinence through different mechanisms.

Conclusions:

The idiographic approach presented provides the potential value of framing recovery as an idiosyncratic dynamic process and provides targets for tailored and adaptive treatment and recovery supporting intervention in future design and evaluation.

Keywords

ecological momentary assessment (EMA)idiographic approach intraindividual variability unified Structural Equation Model (SEM)substance use craving negative affect positive social experiences

Introduction

Conventional longitudinal studies on recovery typically include one baseline or pretest and follow substance abuse patients from the end of treatment for 6 months or a few years but use only a few measurement points (e.g., panel data with three or four waves) during this time period. These studies have provided valuable information on between-individual variation (or interindividual difference) in recovery and have identified reliable trait-like characteristics, such as personality (e.g., McKay & Weiss, 2001), that predict lapse and relapse. Other potential outcome predictors, such as individuals’ levels of negative affect, vary both between and within individuals (i.e., intraindividual variability) substantially through recovery (McKay, Franklin, Patapis, Lynch, 2006; Shiffman, 2009) and have been termed “state like.” Because conventional recovery studies use the nomothetic approach—directly aggregating data from all individuals and generalizing pooled results to the entire population and/or across time and focusing on between-individual analyses (e.g., analysis of variance or conventional regression)—these conventional studies mostly target between-individual variation, making it difficult to soundly draw inferences about intraindividual recovery process (Molenaar, 2004; Molenaar & Campbell, 2009).

As a step toward integrating within-person analyses into recovery research, this article proposes an alternative theoretical framework for investigating and understanding substance abuse recovery as a dynamic daily process. We propose an idiographic perspective that emphasizes the idiosyncratic nature of intraindividual process and cautions against directly aggregating all individuals’ data. This perspective is married to a specific two-step idiographic approach, whereby analyses first focus on intraindividual variability using within-individual analyses (e.g., single-subject time series analysis) to examine individual-level process across time and second identify subgroups with similar intraindividual patterns. This framework is better regarded as a complement rather than a rival of the conventional nomothetic framework. With the aid of ecological momentary assessment (EMA; Shiffman, 2009; Shiffman, Stone, & Hufford, 2008), this article elaborates the following two aspects of this framework: (1) although the majority of conventional studies have focused on a relatively macro-time level (e.g., months and years), it is equally important to understand recovery at the micro daily level—“one day at a time” and (2) although nearly all recovery research has taken a nomothetic approach at the macro-population level, EMA data enable us to examine recovery process at an idiographic within-individual level.

We start the literature review with pertinent recovery research that applies conventional time scales. Next, we set out differences between the between-individual and within-individual analytic frameworks. We then elaborate on the nomothetic and idiographic perspectives. Next, we provide a demonstration of the potential utility of the idiographic approach by applying it to daily diary data provided by a group of abstaining young adults in sustained recovery from substance abuse. The specific analyses model within- and between-day associations among negative affect, substance use craving, and positive social experiences. Although the potential meaning of the demonstration results for understanding recovery process is briefly considered, our primary focus is on the utility of the idiographic framework for recovery research and its potential for informing the design and evaluation of intervention for substance abuse recovery.

Conventional Longitudinal Studies in Recovery

Conventional longitudinal studies have shown that obtaining full recovery from substance abuse can be extremely difficult. In fact, most individuals who go through substance abuse treatment will relapse back into addictive use within the first year or even 6 months after leaving treatment (Dimeff & Marlatt, 1998; McKay & Weiss, 2001; Scott, Foss, & Dennis, 2005). It can take years for recovering individuals to become fully abstinent and attain a sustained recovery (Scott et al., 2005; Weisner, Matzger, & Kaskutas, 2003). Although they provide useful information on the timing, intensity, and correlates of lapse (the initial use after maintaining abstinence for a period) and relapse (“continued use after this initial slip,” Donovan, 1996), conventional studies face two major limitations, that is, (1) they cannot fully assess the intraindividual variability of phenomena related to relapse process and (2) they fail to depict the dynamic process that lead to relapse or sustained abstinence (McKay & Weiss, 2001; McKay et al., 2006; Shiffman, 2009).

A variety of individual-level characteristics have been associated with lapse and relapse. Some of these characteristics, such as personality (e.g., neuroticism), can be regarded as trait-like variables that are generally stable across time within individuals (i.e., large interindividual difference, but little intraindividual variability). One possible research question involving these trait-like variables is, Will individuals with low level of problem solving coping be more likely to relapse than individuals with high level of problem solving coping? Conventional studies and between-individual analyses are well suited to address this type of research question.

Many psychological constructs, such as negative affect and craving, beyond demonstrating interindividual differences, also demonstrate substantial intraindividual variability (Donovan, 1996; Witkiewitz & Bowen, 2010). These variables would be better viewed as state-like variables. Given the significant covariation between these variables and recovery outcomes on a between-individual level, we suggest it is critical to consider intraindividual variability in these constructs (McKay et al., 2006; Ridenour, Meyer-Chilenski, & Reid, 2012; Shiffman, 2009; Witkiewitz & Bowen, 2010). A possible research question involving these state-like variables would be is an individual on a day with particular high level of negative affect more likely to relapse than he or she is on a day with low level of negative affect?

However, given the sparse measurement points typically provided by most panel designs (i.e., assessing individuals every few months), conventional longitudinal studies do not capture sufficient intraindividual variability in these state-like variables. As a result, these designs are unable to examine the dynamic daily process that helps maintain abstinence or, alternatively, leads to relapse. Besides focusing on how many people relapse after how much time and what individual-level characteristics predict relapse, recovery research should also focus on understanding patterns of within- and between-day associations (Donovan, 1996; Shiffman, 2009; Witkiewitz & Marlatt, 2004). What happens in the dynamic process from the time individuals leave treatment until they establish a sustained recovery or relapse? Intensive longitudinal assessments such as EMA are needed to sufficiently capture such intraindividual variability (McKay et al., 2006; Shiffman, 2009) and to answer these questions.

Between-Individual Versus Within-Individual Analyses

The majority of substance abuse recovery studies are based on between-individual analyses of a sample meant to generalize to a population, with conclusions based on the interindividual difference in individual-level characteristics and outcomes in the sample. For example, individuals with above-average negative affect are more likely to relapse after a year than individuals with below-average negative affect (McKay, 1999; Witkiewitz & Bowen, 2010). Generalizing from conclusions based on the interindividual difference and between-individual analyses to intraindividual process, however, is problematic (Molenaar, 2004; Molenaar & Campbell, 2009). Such inferences can only be based on the within-individual analyses.

The classic ergodic theorem demonstrates that only under rare occasions—the homogeneity of the population and the stationarity of psychological or behavioral process—do between-individual analyses yield the same results as within-individual analyses (Molenaar, 2004; Molenaar & Campbell, 2009). The population homogeneity assumption requires that all recovering individuals follow the same or rather similar process. This assumption is unlikely to be true, especially over the long term. First, interindividual heterogeneity is quite common (Scott et al., 2005; McKay et al., 2006; Witkiewitz, van der Maas, Hufford, & Marlatt, 2007). Second, the developmental principle of equifinality suggests that individuals can relapse through different paths (Cicchetti & Rogosch, 1996). Thus, recovery processes are unlikely to be homogeneous.

The stationarity assumption states that both the mean and the variation in psychological or behavioral constructs remain constant across time. It is difficult to believe such stationarity can occur in processes that contribute to relapse or abstinence maintenance. For example, for those who do relapse, most will first lapse (use substance at a lower level or frequency than typical of their pretreatment use) before fully relapse (return to the same use intensity as pretreatment; Donovan, 1996; Marlatt & Gordon, 1985), therefore suggesting a process of gradually increasing risk of relapse. Moreover, lapse or relapse may also be preceded or accompanied by an increase in negative affect or other psychosocial phenomena. Collectively, these two assumptions emphasize the necessity of within-individual analyses for understanding intraindividual processes in recovery research.

Nomothetic Versus Idiographic Approaches

Besides focusing on a macro-time scale level, most recovery studies also adopt the nomothetic approach. The nomothetic approach aggregates data across all individuals and generalizes the averaged results to an entire population. Despite the wide use and apparent generalizability of results, critics have claimed that such pooled results may not fit any specific individual at all (Hoeppner, Goodwin, Velicer, Mooney, & Hatsukami, 2008; Molenaar, 2004; Ridenour, Pineo, Maldonado, & Hassmiller, 2013). The difficulty in applying results from nomothetic analyses to any specific individual has led some to focus on developing idiographic approach to behavioral and social science.

Outside the recovery area, the population homogeneity assumption has been undercut by many empirical studies using mixture methods (e.g., Muthén & Shedden, 1999), which focus on identifying homogeneous subgroups in the population. The focus on population subgroups is consistent with within-individual analyses and idiographic approach to understanding recovery, which focuses on interindividual difference in intraindividual variability (Witkiewitz & Marlatt, 2004). Applying these approaches allows for detecting distinct subgroups that each demonstrates different recovery patterns and shows substantial similarity of patterns among individuals in the same subgroup (e.g., Witkiewitz et al., 2007). A relevant research question under this approach would be is negative affect differently associated with relapse across subgroups? It may be a prominent trigger in one subgroup but irrelevant in another.

Addressing these types of research questions for understanding recovery ultimately requires a framework for investigating intraindividual variability and interindividual difference in intraindividual variability as well as a matching analytic approach that closely investigates patterns of risk and protective factors over time (Donovan, 1996; Marlatt & Gordon, 1985; Witkiewitz & Marlatt, 2004). The proposed idiographic analytic approach first focuses on individual-level data, capturing each individual’s unique intraindividual recovery process. Next, analyses cluster individuals into homogenous subgroups, revealing distinct recovery patterns that reflect interindividual difference in intraindividual variability.

Empirical Recovery Studies Adopting EMA and the Current Study

A few empirical studies applying EMA to alcohol and drug abuse recovery have been published. Cleveland and Harris (2010) used multilevel modeling to investigate the moderation effects of different coping strategies (e.g., problem solving) on the within-day associations between negative affect, negative social interactions (e.g., hostility and insensitivity), and same-day cravings. This study demonstrated substantial intraindividual variability in these state-like variables. EMA has been more often applied to the study of smoking relapse than to alcohol and other drug abuse recovery (McKay et al., 2006; Shiffman, 2009). Shiyko, Lanza, Tan, Li, and Shiffman (2012) applied a time-varying effect model to examine the dynamic association between negative affect and self-efficacy on smoking urges during the 2 weeks following a quit attempt comparing within-day associations between successful quitters and relapsers who demonstrated different recovery/relapse patterns. Both Cleveland and Harris (2010) and Shiyko et al. (2012), however, only examined within-day associations (or contemporaneous relationships) and single outcome. In contrast to modeling within-day associations, Hoeppner, Goodwin, Velicer, Mooney, and Hatsukami (2008) used time series and dynamic cluster analysis to identify three distinct subgroups that demonstrated different between-day patterns following a posttreatment smoking cessation sample, namely, decreasing, constant, and increasing. This study adopted an idiographic approach to recovery process and also revealed individual-level characteristics that predicted subgroup membership. Zheng, Wiebe, Cleveland, Molenaar, and Harris (2013) extended this research into multivariate between-day associations (or sequential relationships) using vector autoregression (VAR) to examine associations between negative affect, craving, and smoking. Specifically, Zheng et al. (2013) applied an idiographic approach by first fitting individual process models, then identifying subgroups showing distinct recovery processes.

Building on the earlier studies, this article calls for an alternative theoretical framework that considers and examines recovery as an idiographic dynamic daily process. To demonstrate the feasibility and potential importance of this approach, this study investigates both within- and between-day associations (collectively day-to-day associations) among substance use craving, negative affect, and positive social experiences in a group of abstaining young adults in sustained recovery from alcohol and drug use. These youth were members of a Collegiate Recovery Center (CRC). CRCs are largely based on the 12-step program, participating in which has been shown to be effective in reducing relapse (Humphreys, Mankowski, Moos, & Finney, 1999). These centers provide comprehensive recovery support services.

Given that no relapse occurred during the span of data collection, variables of interests were chosen to demonstrate how risk and protective processes of abstinence maintenance can be understood at a daily level. Negative affect and craving were selected because they have been consistently shown as risk factors for relapse (Marlatt & Gordon, 1985; McKay, 1999; McKay & Weiss, 2001; Witkiewitz & Bowen, 2010). In addition, craving may be regarded as a cognitive state that can relieve the immediate feeling of negative affect and trigger relapse by interfering with the negative reinforcement process through which recovery is sustained (see Baker, Piper, McCarthy, Majeskie, & Fiore, 2004; Robinson, Lam, Carter, Wetter, & Cinciripini, 2012). Thus, a positive between-day association between negative affect and cravings may exist as found in one subgroup in Zheng et al. (2013). Positive social experiences have been identified as protective factor against relapse, possibly by mitigating enhanced level of negative affect and/or craving (Dobkin, Civita, Paraherakis, & Gill, 2002; Humphreys, Moos, & Cohen, 1997; Moos & Moos, 2007), suggesting a negative between-day association of positive social experiences with these two risk variables.

The analyses were conducted as follows. In the first step, in order to investigate patterns among the three variables, first-order unified Structural Equation Models (SEMs), that is, unified SEM(1), modeling both contemporaneous and sequential relationships (see Gates, Molenaar, Hillary, Ram, & Rovine, 2010; Kim, Zhu, Chang, Bentler, & Ernst, 2007, for detailed mathematical representations) were separately applied to each individual. Unified SEM(1) was chosen over VAR(1) as used in Zheng et al. (2013) because it can also model within-day associations. Unified SEM(1) belongs to the family of time series analysis and could be considered as cross-lagged model in a time series scenario. We expect that these models will reveal substantial heterogeneous patterns across individuals, whereby within- and between-day associations will differ in both direction and magnitude. Second, individual unified SEM(1) estimates were then entered into cluster analyses to identify homogenous subgroups. Third, once subgroups were identified, data were aggregated at the subgroup level by pooling across individuals in the same cluster, with the expectation that distinct subgroups will demonstrate different processes with interpretable meanings. The goal of the empirical demonstration is to present a concrete example of the feasibility and value of considering interindividual heterogeneity when investigating intraindividual recovery processes. Given the exploratory nature and demonstrative purpose of the analyses, no specific predictions were made regarding the number or nature of identified homogenous subgroups.

Method

Sample

The full sample consisted of 55 young adults (female = 16, mean age = 22.6, SD = 5.8), active members of a CRC at a Southwestern university, who provided daily diary data once per day with personal digital assistant (PDA) devices at the end of each day (see Cleveland & Harris, 2010, for details on study design, participant recruitment, and data collection). For the current analysis, the final sample included 39 young adult (female = 12, mean age = 22.9, SD = 6.3). Three were excluded because their craving exhibited no intraindividual variability. Simulation study suggested that data less than 20 days and/or with too many missing values caused model identification problems and biased estimates (Velicer & Colby, 2005). Ten more participants provided less than 20 days of valid data, seven of whom also provided more than 5 days of missing data (25% missing values vs. 7.1% in the final sample) and thus were excluded. Preliminary analyses indicated that 3 of the remaining 42 could not be fit to a satisfactory unified SEM(1), which suggested that 1-day lag model could not sufficiently explain dynamic association and required higher order models. Such higher order models went beyond the capability of available data and the scope of this study, thus were excluded.

In the final sample, all participants were non-Hispanic Whites; 35 (90%) had received professional alcohol/drug dependency treatment; 28 (72%) had received inpatient care or addiction treatment, for 3 months or more in nearly all cases. All considered themselves to be 12-step group members and reported reading 12-step literature and applying the steps to their lives on a daily basis. Participants provided an average of 26.7 days’ worth of data (SD = 3.1, ranging from 22 to 35), with an average of 1.9 missing days (SD = 2.1, ranging from 0 to 8). The 16 excluded participants (female = 4, mean age = 21.9, SD = 4.3) showed similar demographic characteristics and treatment history, that is, 15 were non-Hispanic Whites, all had received treatment, and 9 had received inpatient care or addiction treatment. In addition, their mean levels and intraindividual variability of the three time-varying variables did not differ from those in the final sample, for example, 1.68 (SD = 0.38) in excluded sample versus 1.65 (SD = 0.42) in final sample for negative affect.

Measures

Substance use craving

Daily substance use craving was measured with 7 items modified from the Desires for Alcohol Scale (Love, James, & Willner, 1998) and the Alcohol Urges Questionnaire (Bohn, Krahn, & Staehler, 1995) to accommodate daily assessment and polydrug use. A sample item reads, “For a moment today I missed the feeling of drinking or drugging.” Responses ranged from 1 = strongly disagree to 5 = strongly agree. Cronbach’s α was calculated both conventionally, without regard to reports being nested within participants, and separately for each analysis day. Conventional α was .95, while within-day α ranged from .88 to .97.

Negative affect

The 10-item scale from the Positive and Negative Affect Scale (PANAS; Watson, Clark, & Tellegen, 1988) was used to assess daily negative affect. Emotions included “stressed,” “upset,” “scared,” “hostile,” and “irritable.” Responses ranged from 1 = very slightly or not at all to 5 = very much. Conventional α was .88, while day-level α ranged from .78 to .93.

Positive social experiences

This construct was measured by a 5-item scale modified from the Test of Negative Social Exchange (TENSE) scale to capture positive social interactions (Ruehlman & Karoly, 1991), where participants responded 1 = Yes or 0 = No, if any of the following happened today: Did anyone show you that they cared about you today? Did anyone offer to help you with a problem today? Did anyone give you advice on something today? Did anyone share their point of view about a problem you were having today? Did anyone show any concern for you today? Conventional α was 0.83, while day-level α ranged from .69 to .95.

Analytic Strategy

Three variables of interests were first individually regressed on time to detect and remove any potential linear trends in preliminary analysis. Few participants exhibited mild linear trends, mostly for negative affect (see Table 1). Block Toeplitz covariance matrices of up to 2 days’ lag from multivariate time series data of each participant were calculated, with missing data pairwise deleted when producing the Block Toeplitz covariance matrix (Molenaar, 1985). Next, unified SEM(1) was fit to these covariance matrices for each participant. As specified in Figure 1, daily craving (C₂) was predicted by prior day’s craving (C₁; autoregression) as well as prior day’s negative affect (N₁) and positive social experiences (P₁; cross-lagged). In addition, it was also predicted by the same day’s positive social experiences (P₂). The same procedure was applied to daily negative affect (N₂). Predictions of positive social experiences by the previous day’s three variables were not included because we were mostly interested in the way positive social experiences reduce, eliminate, or exacerbate recovery risks imposed by the cross-lagged and autocorrelation of cravings and negative affect. We used LISREL 8.8 (Jöreskog & Sörbom, 2006) with quasi-maximum likelihood estimation, which has been shown to produce the same unbiased estimates as full information maximum likelihood estimation using Block Toeplitz covariance matrix (Hamaker, Dolan, & Molenaar, 2005). To protect the .05 α level at individual model level and to avoid spurious significance, Bonferroni correction was used to adjust for multiple tests within each individual unified SEM(1), producing an adjusted single-test α level of .006 (.05/8).

Figure 1.

Conceptual model of unified first order SEM. N = negative affect; C = craving; P = positive social experiences. Regression residuals for Time 2 variables omitted from figure.

Table 1.

Individual Means (SDs) and Linear Trends of Substance Use Craving, Negative Affect, and Positive Social Experiences.

ID	No.	Mean (SD)			Linear Slope			ID	No.	Mean (SD)			Linear Slope
ID	No.	N	C	P	N	C	P	ID	No.	N	C	P	N	C	P
1	23	1.44 (0.46)	1.08 (0.37)	0.68 (0.30)	0.01	0.01	0.03*	21	22	1.25 (0.40)	1.48 (0.88)	0.54 (0.18)	0.01	0.02	−0.00
2	26	1.63 (0.49)	1.16 (0.44)	0.73 (0.29)	0.02	0.01	0.01	22	27	1.41 (0.44)	1.27 (0.87)	0.59 (0.23)	−0.01	−0.04	−0.01
3	28	1.91 (0.39)	1.47 (0.84)	0.89 (0.19)	0.01	−0.01	0.01	23	25	1.83 (0.76)	1.14 (0.26)	0.69 (0.32)	−0.00	0.01	0.02
4	27	1.41 (0.54)	2.54 (1.12)	0.58 (0.29)	−0.04*	−0.10*	−0.03*	24	26	1.77 (0.57)	1.37 (0.92)	0.90 (0.22)	−0.03	−0.01	0.00
5	22	2.54 (0.41)	1.13 (0.28)	0.88 (0.22)	0.02	0.01	−0.00	25	24	1.98 (0.37)	1.12 (0.39)	0.84 (0.30)	−0.00	0.01	0.01
6	30	1.29 (0.29)	1.48 (0.72)	0.49 (0.21)	−0.00	−0.01	−0.00	26	25	2.61 (0.42)	2.13 (1.14)	0.63 (0.29)	0.02	0.03	−0.00
7	25	1.39 (0.51)	1.32 (0.58)	0.95 (0.13)	−0.01	−0.01	0.01	27	24	1.29 (0.26)	1.05 (0.13)	0.78 (0.26)	0.00	0.00	0.01
8	25	1.34 (0.39)	1.09 (0.31)	0.31 (0.40)	0.00	0.01	0.02	28	22	1.66 (0.82)	1.81 (1.21)	0.88 (0.25)	0.05	−0.01	0.01
9	25	1.29 (0.35)	1.13 (0.28)	0.90 (0.25)	0.00	−0.01	0.01	29	26	1.64 (0.48)	1.10 (0.20)	0.44 (0.32)	0.02	−0.00	0.01
10	26	2.75 (0.61)	1.92 (1.08)	0.57 (0.34)	−0.04*	−0.01	−0.02	30	30	1.32 (0.19)	1.04 (0.14)	0.19 (0.21)	−0.00	−0.01	−0.01
11	25	1.44 (0.58)	1.25 (0.77)	0.50 (0.31)	0.00	−0.01	−0.01	31	26	1.49 (0.26)	1.56 (0.67)	0.85 (0.24)	0.02*	0.03	0.00
12	29	1.79 (1.00)	1.69 (1.25)	0.85 (0.30)	−0.04	−0.03	−0.02	32	31	1.22 (0.34)	1.47 (0.89)	0.28 (0.34)	−0.01	−0.02	−0.01
13	25	1.36 (0.52)	1.19 (0.66)	0.22 (0.32)	−0.03	−0.04	−0.01	33	25	1.47 (0.52)	1.41 (0.52)	0.10 (0.17)	−0.01	−0.03	−0.02*
14	24	1.26 (0.28)	1.82 (1.19)	0.81 (0.29)	−0.01	0.06	−0.01	34	28	2.20 (0.86)	2.67 (1.64)	0.65 (0.40)	0.06*	0.13*	−0.00
15	28	1.34 (0.22)	1.12 (0.24)	0.35 (0.22)	0.00	−0.01	0.01	35	33	2.73 (0.93)	2.81 (1.44)	0.33 (0.41)	0.03	0.02	−0.01
16	35	1.65 (0.36)	2.08 (0.88)	0.51 (0.31)	−0.00	−0.01	−0.00	36	28	1.80 (0.77)	1.72 (1.00)	0.37 (0.34)	0.01	−0.02	0.01
17	31	1.94 (0.66)	1.81 (1.07)	0.40 (0.37)	−0.01	0.01	−0.01	37	28	1.29 (0.34)	1.04 (0.08)	0.96 (0.20)	−0.01	−0.00	0.00
18	29	1.36 (0.56)	1.38 (0.75)	0.67 (0.21)	0.01	−0.01	−0.01	38	33	1.60 (0.59)	1.47 (0.89)	0.39 (0.30)	−0.02	−0.01	−0.01
19	27	1.65 (0.42)	2.35 (0.79)	0.60 (0.30)	0.03*	0.02	−0.00	39	26	1.33 (0.29)	1.31 (0.65)	0.76 (0.27)	−0.02	0.00	0.01
20	24	1.53 (0.40)	1.13 (0.50)	0.45 (0.30)	−0.01	−0.01	0.00

Note. N = negative affect; C = craving; P = positive social experiences; No. = number of observations of each participant. ID is randomly assigned to each individual.

*p < .01.

After fitting individual unified SEM(1), individual model’s estimates were used in cluster analysis to identify homogenous subgroups. Model estimates rather than the measured variables were used because we were interested in interindividual difference of intraindividual patterns, which would manifest in different patterns of day-to-day associations of variables. Given the limited number of participants as well as the exploratory nature of this analysis, we used a two-step auto-cluster analysis in SPSS 17.0 (SPSS, 2001). Individuals with similar model estimates were first preclustered into subclusters by constructing a modified cluster feature tree. Next, using the generally sensitive log-likelihood distance measure (the corresponding decrease in log-likelihood from combining two clusters into one) and Akaike’s Information Criterion, numbers of clusters were automatically determined and individuals with similar model estimates were grouped into clusters using the agglomerative hierarchical clustering method. In the last step, using multigroup unified SEM(1), model estimates were reestimated using the same specification described earlier, and pooled data across all participants in the same cluster to produce cluster-level estimates. Specifically, regression estimates in the same cluster were constrained to be equal across all participants, whereas individual contemporaneous correlations of variables and residuals were left to be freely estimated.

Results

Individual Unified SEM(1)

Fit indices for all individual unified SEM(1) indicated largely satisfactory fit. Table 2 summarized the standardized estimates of each individual, model fit indices, and the clusters into which the individuals were grouped. As shown, individuals exhibited substantial heterogeneity in their day-to-day associations of negative affect, craving, and positive social experiences. Five participants’ negative affect showed significant autoregression, two of which were negative and all fell in Cluster 2. Four participants’ negative affect predicted next day craving, three of which were positive and all fell in Cluster 1. Three participants’ craving significantly predicted next day negative affect positively, and all fell in Cluster 2. Four participants’ craving exhibited significant autoregression, all positive and three were in Cluster 1.

Table 2.

Model Fit and Parameter Estimates of Individual First-Order Unified SEM (Unified SEM(1)) by Clusters.

ID	Unified SEM(1) Model Fit Indices				Standardized Unified SEM(1) Estimates
	Error Variance			χ²/df	N₂ (DV)				C₂ (DV)
	N	C	P	χ²/df	N₁	C₁	P₁	P₂	N₁	C₁	P₁	P₂
Cluster 2
1	0.16	0.09	0.08	0.51	0.04	0.26	−0.39^†	0.42^†	−0.33	0.24	−0.47*	0.52*
2	0.19	0.16	0.08	0.43	−0.19	−0.03	0.16	0.34^†	−0.16	−0.07	−0.18	−0.15
3	0.12	0.53	0.04	0.47	−0.01	−0.05	0.33^†	0.33^†	−0.29^†	0.15	0.26^†	0.32^†
4	0.24	0.45	0.07	0.60	−0.03	0.27	−0.03	0.20	−0.01	0.71*	−0.34*	0.43*
5	0.14	0.07	0.05	0.56	0.00	0.26	−0.13	0.14	−0.13	0.02	0.03	−0.01
6	0.06	0.44	0.04	0.93	−0.13	−0.20	0.13	0.39*	−0.14	0.28^†	0.17	0.11
7	0.22	0.32	0.02	0.20	−0.05	0.37	0.05	0.06	0.03	−0.04	0.09	−0.02
8	0.10	0.08	0.15	1.12	−0.34*	0.15	−0.16	0.65	0.29	−0.05	−0.19	0.02
9	0.09	0.07	0.06	0.50	0.06	0.24	0.32^†	0.35^†	0.05	−0.06	0.15	−0.15
10	0.21	1.03	0.11	0.14	0.22	0.37*	0.20	0.44*	0.16	−0.20	0.18	−0.22
11	0.28	0.49	0.09	0.49	0.05	0.03	0.28^†	0.16	−0.25	0.09	0.26	0.23
12	0.76	1.33	0.09	0.31	−0.19	0.30^†	0.13	0.34^†	−0.31^†	0.20	0.03	0.14
13	0.21	0.41	0.10	0.68	−0.14	0.47*	−0.02	0.07	−0.11	0.03	0.04	0.12
14	0.05	1.06	0.08	0.34	−0.22	0.17	0.32^†	0.55*	−0.05	0.18	0.08	0.46*
15	0.04	0.05	0.05	0.65	−0.17	0.36^†	0.12	0.08	−0.25	−0.03	0.14	0.04
16	0.07	0.58	0.09	0.96	−0.47*	0.01	0.40*	0.52*	−0.56*	0.12	0.23^†	0.40*
17	0.30	1.08	0.13	0.98	−0.28^†	0.45*	0.41*	0.53*	−0.03	−0.02	−0.05	−0.02
Cluster 1
18	0.27	0.40	0.04	0.38	0.23	−0.07	−0.17	0.26	0.09	−0.14	−0.49*	−0.20
19	0.12	0.44	0.09	0.57	0.56*	−0.03	−0.20	0.09	0.44*	−0.01	−0.22	0.04
20	0.13	0.22	0.08	1.11	0.03^†	−0.23	−0.47*	0.37^†	−0.11	−0.17	−0.07	−0.04
21	0.13	0.68	0.03	0.28	−0.06	0.06	−0.12	0.34	0.23	−0.33	−0.04	−0.02
22	0.15	0.65	0.05	0.78	0.21	−0.17	−0.26	0.28^†	0.30^†	−0.11	−0.09	0.03
23	0.49	0.03	0.11	0.80	0.32	−0.04	0.03	0.04	0.38*	0.52*	−0.04	−0.09
24	0.20	0.73	0.04	0.48	0.27^†	−0.04	−0.56*	0.12	0.21	−0.23	0.05	0.10
25	0.11	0.14	0.09	0.36	0.33^†	−0.02	−0.20	−0.06	0.04	−0.10	0.13	−0.21
26	0.08	1.10	0.08	0.39	0.57*	0.16	−0.04	0.35*	−0.01	−0.08	−0.33^†	−0.03
27	0.05	0.01	0.07	0.77	0.14	−0.21	−0.14	0.30	0.75*	−0.16	−0.23	−0.15
28	0.52	1.30	0.07	1.18	0.41^†	0.17	−0.05	0.13	0.36^†	−0.17	−0.25	0.09
29	0.14	0.04	0.10	0.23	0.19	−0.25^†	0.14	0.39*	0.14	−0.07	−0.34^†	0.20
30	0.03	0.01	0.04	0.58	0.06	−0.09	0.11	−0.04	0.14	0.53*	−0.01	0.16
31	0.05	0.38	0.06	0.28	0.30	0.11	−0.04	0.27	0.18	0.08	0.10	0.14
32	0.10	0.68	0.11	0.53	0.16	−0.11	−0.20	0.28^†	0.12	−0.39^†	0.04	−0.05
33	0.22	0.19	0.03	0.95	0.26	−0.03	−0.33^†	0.17	0.14	−0.14	0.45*	0.13
34	0.42	0.50	0.16	1.22	0.38^†	0.26	−0.38*	0.34^†	0.19^†	0.73*	−0.55*	0.15^†
35	0.81	1.58	0.16	0.25	0.12	−0.20	−0.11	0.08	−0.10	−0.06	0.42*	−0.22
36	0.39	0.50	0.11	1.47	0.52^†	0.00	−0.07	−0.08	0.18	0.49^†	0.04	−0.08
37	0.10	0.01	0.04	0.62	0.08	−0.18	0.13	0.19	0.08	−0.24	0.15	0.13
38	0.26	0.72	0.09	0.47	0.25	−0.28^†	0.06	0.30^†	−0.05	−0.03	0.19	0.08
39	0.06	0.35	0.07	0.65	0.45*	−0.11	−0.13	0.18	0.27	−0.05	0.16	−0.16

Note. N = negative affect; C = craving; P = positive social experiences; DV = dependent variable. The same ID indicates the same individual as in Table 1. All but one model (p = .07) had a p value higher than .21. All but one model (comparative fit index [CFI] = .88) had a CFI higher than .91. All but one model (non-normed fit index [NNFI] = .83) had a NNFI higher than .87. All models had negligible root mean square error of approximation [RMSEA], with the five highest RMSEA value being .068, .07, .091, .09, and .13, respectively (conventionally below .05 as satisfactory, and between .05 and .10 as acceptable). However, the other model fitting indices were satisfactory and thus these models were retained.

^† p < .05. *p < .006.

Five participants’ positive social experiences significantly predicted next day negative affect, three being negative and all in Cluster 1. Six participants’ positive social experiences predicted next day craving, two of which were positive and all in Cluster 1. Regarding within-day associations, seven participants’ positive social experiences were positively associated with the same day negative affect and fell in both clusters roughly even. Four participants’ positive social experiences were also positively associated with the same day craving, all in Cluster 2.

Cluster Analysis

As indicated in Table 2, two distinct clusters were identified, with 22 belonging to Cluster 1 and 17 to Cluster 2. The existence of some specific estimates in only one cluster (e.g., all negative associations between positive social experiences and next day negative affect in Cluster 1, whereas all positive associations in Cluster 2) support the idea that within-cluster individual patterns were substantively similar. Cluster-level estimates were shown in Table 3. The two clusters were more different than similar, suggesting distinct day-to-day recovery processes. First, negative affect demonstrated positive autoregression in Cluster 1 (.25) but not in Cluster 2 (−.06, not significant [ns]). Craving showed positive autoregression at non-Bonferroni-corrected α level in both clusters (both .08). Second, negative affect predicted next day’s craving in both clusters but in different directions (.09 and −.10). Craving predicted next day’s negative affect in Cluster 2 (.10) but not in Cluster 1 (.00, ns). Third, positive social experiences were positively associated with the same day’s negative affect in both clusters (.14 vs. .20). They were not associated with the same day’s craving in either cluster (.00 and .04, ns), although the prediction in Cluster 2 was significant at non-Bonferroni-corrected α level. They did not predict the next day’s craving either (−.01 and .02, ns); however, they predicted next day negative affect in both clusters but in different directions (−.10 and .09).

Table 3.

Patterns of Day-to-Day Association in Clusters.

IV	DV	Cluster 1 (n = 22)			Cluster 2 (n = 17)
IV	DV	B	SE	β	B	SE	β
Negative affect 1	Negative affect 2	.25*	.03	.25*	−.06	.03	−.06
Craving 1	Negative affect 2	.00	.02	.00	.06*	.02	.10*
Positive social experiences 1	Negative affect 2	−.19*	.04	−.10*	.16*	.05	.09*
Positive social experiences 2	Negative affect 2	.27*	.04	.14*	.36*	.04	.20*
Negative Affect 1	Craving 2	.14*	.02	.09*	−.16*	.04	−.10*
Craving 1	Craving 2	.08^†	.03	.08^†	.08^†	.03	.08^†
Positive social experiences 1	Craving 2	−.04	.03	−.01	.06	.06	.02
Positive social experiences 2	Craving 2	.01	.03	.00	.12^†	.06	.04^†

Note. IV = independent variable; DV = dependent variable; B = unstandardized regression coefficient; β = standardized regression coefficient. Negative affect 1: negative affect at time 1; craving 2: craving at time 2. The same applies to all other variables.

^† p < .05. *p < .006.

To investigate individual-level characteristics that could differentiate the two clusters from each other, a series of t-tests were conducted on several time-invariant variables. These included past alcohol use and drug use experiences, alcohol anonymous self-efficacy, support-seeking coping, problem-solving coping, avoidance coping as well as friend and family support. No significant difference existed between these two clusters in any of these variables. In addition, cluster means, as well as the average of intraindividual variability of the three time-varying variables, were also compared between clusters and revealed no significant difference. In other words, the two clusters did not differ in the means of interindividual characteristics we examined or in means and intraindividual variability of the three time-varying variables.

Discussion

Sustained recovery from substance abuse happens on a day-to-day basis. This intraindividual process involves the dynamic interactions of risk and protective factors (Donovan, 1996; Witkiewitz & Marlatt, 2004). Many state-like variables, such as negative affect and craving, demonstrate considerable intraindividual variability (Ridenour et al., 2012). Such variability requires appropriate data collection methods such as EMA (Shiffman, 2009) to capture the complexity of intraindividual process and corresponding analytic techniques that can properly model this process (McKay et al., 2006). This article proposed an alternative theoretical framework for examining substance abuse recovery as an idiographic process at a daily and individual level. To demonstrate the feasibility and capability of this approach, this study applied unified SEM(1) to investigate day-to-day associations among positive social experiences, craving, and negative affect in a CRC sample of young adults. Individual unified SEM(1) revealed considerable interindividual heterogeneity in their day-to-day association patterns. Cluster analysis identified two subgroups with different recovery processes.

Interindividual Heterogeneity Revealed by Idiographic Intraindividual Analyses

In individual unified SEM(1), both craving and negative affect showed significant autocorrelations. These autocorrelations existed only among some participants and in both directions. Most autocorrelations were positive, suggesting that higher craving and negative affect can persist to the next day. These positive between-day associations may create opportunity for cascades, whereby accumulated craving and negative affect can impose difficulty in abstinence maintenance without proper and in-time treatment (Marlatt & Gordon, 1985; Witkiewitz & Marlatt, 2004). It should be emphasized, however, that a substantial proportion of participants showed no such autocorrelation (see Table 2). Some participants also demonstrated reciprocal influence of negative affect and craving. Most of these associations were positive, suggesting that an increase in one variable could lead to an increase in another the next day (Witkiewitz & Marlatt, 2004; Zheng, Wiebe, Cleveland, Molenaar, & Harris, 2013), which may also challenge recovery stability.

Individual unified SEM(1) also revealed significant associations of positive social experience with craving and negative affect, both within- and between-day. A substantial proportion of participants’ positive social experiences predicted next day’s craving and negative affect in both directions. Regarding within-day associations, many individuals’ high level of positive social experiences was associated with high levels of the same day’s negative affect and craving. However, no causal conclusion should be made regarding this specific within-day association, which could be explained by individuals actively seeking social support from other CRC members when experiencing negative affect.

Subgroups with Distinct Recovery Processes Revealed by Cluster Analysis

Considering clusters separately provides useful insights about the heterogeneity in day-to-day process among young adults in sustained recovery. In Cluster 1, both increased negative affect and craving could persist to the next day. In addition, enhanced negative affect could also lead to enhanced craving in the next day (Witkiewitz & Marlatt, 2004). Therefore, craving appeared to be an outcome of negative affect among individuals in this cluster, who possibly increased their substance use cravings as a way to relieve the immediate feelings of negative affect (Baker et al., 2004; Robinson et al., 2012). It appears that positive social experiences could help to maintain recovery by mitigating increased negative affect, but not craving. Higher level of positive social experiences was associated with higher level of same day negative affect as well as lower level of next day negative affect. It may be that, given negative affect’s risk to recovery sustainability, these participants purposely seek positive experiences and support the day they experience higher than normal level of negative affect, resulting in lower negative affect level the next day. The negative association between positive social experiences and the next day’s negative affect is consistent with the presumed role of social support in helping maintain recovery (Dobkin et al., 2002; Humphreys et al., 1997; Moos & Moos, 2007).

The situation differed in Cluster 2. There was no autocorrelation for negative affect. In addition, there were between-day reciprocal associations between craving and negative affect but in different directions. Therefore, for individuals in Cluster 2, increased craving could lead to enhanced negative affect the next day, which could predict decreased craving the day after. In this cluster, it appears that negative affect was an outcome of craving. This negative feedback pathway provides an example of how stable levels of psychosocial states can be maintained in dynamic process over time (Marlatt & Gordon, 1985; Witkiewitz & Marlatt, 2004).

The positive association between positive social experiences and the next day’s negative affect in Cluster 2 seems to be counterintuitive. However, Cluster 2 also differed from Cluster 1 in its negative association between negative affect and the next day’s craving. When considered across 3 days, the combined impact of these two paths created a negative association between positive social experiences and craving 2 days later. Therefore, although the variables of interest fit satisfactorily to a first-order model, where their effects could carry over to the next day, the collective influences of these variables in maintaining sustained recovery need to be understood across multiple days (see also Zheng et al., 2013). Alternatively, this positive association in Cluster 2 may also indicate a potential iatrogenic effect for this particular subgroup. Although the specific findings regarding the different clusters are intriguing, we suggest these findings be viewed as evidence as to the capability of idiographic approach and analyses to capture interindividual heterogeneity in intraindividual processes, rather than substantively interpreted in any more than a suggestive preliminary fashion. Future studies would be needed to replicate and further explore the processes identified in these clusters.

Implications for Understanding and Evaluating Recovery Intervention

Substance abuse remains a major public health issue, especially among young adults attending college (Larimer & Cronce, 2007). Given the rising body of young adults who have gone through substance use treatment and currently in recovery, it is urgent to devote more attention to the day-to-day recovery process at a micro daily and individual level. Results of this alternative approach have implications for those who seek to understand and support these young adults and for future intervention design and evaluation. First, the findings of the interindividual heterogeneity in recovery process demonstrate that the assumption of population homogeneity is not tenable in this case and underscore the need for intraindividual analyses. An idiographic approach is needed to investigate and understand recovery as a dynamic daily process. Appropriate within-individual analyses require data collected through EMA.

Second, rather than assuming that “one size fits all,” potentially different responses to the same intervention warrant close investigation because a given intervention may help some individuals but have iatrogenic effects on others (e.g., Cluster 2; Ridenour et al., 2013; Witkiewitz et al., 2007). Such difference may explain why some people leave treatment and build successful recovery, while others quickly relapse. These differences may not be solely due to difference in patients but to difference in how well individuals are fit to the treatment they receive (Dimeff & Marlatt, 1998; Larimer & Cronce, 2007; McKay & Weiss, 2001). Therefore, using an idiographic approach to identify subgroups with different recovery patterns can provide valuable information for developing and evaluating tailored intervention, where individuals exhibiting different risk factors can be targeted specifically. For example, specific intervention components can be adaptively adopted to reduce or eliminate adverse recovery paths and to reinforce positive ones. In addition, risk and protective factors that demonstrate considerable intraindividual variability at a microlevel can be considered as tailoring variables to maximize effects of long-term time-varying relapse prevention by monitoring individual recovery process closely (Collins, Murphy, & Bierman, 2004; McKay, 2009). For example, relapse prevention programs may increase effectiveness by fostering positive social experiences among members as in Cluster 1.

Limitations and Future Directions

This study is among the first few applying unified SEM with an idiographic approach to investigate day-to-day recovery processes. A few limitations warrant attention. First, choosing participants with more than 3 weeks of valid data and few missing values surely improved the power of individual time series model, but the total sample size was nevertheless small, which limited the capability of cluster analysis. Accordingly, the current results should be considered as demonstrations of the approach and method rather than “substantive findings” per se. In addition, such small sample results, even if substantively correct, would need to be replicated with longer time series, larger samples, and advanced mixture methods and random effects models (e.g., Muthén & Shedden, 1999; Witkiewitz et al., 2007). Second, despite that both contemporaneous and sequential relationships of positive social experiences with other variables were considered, no causal conclusions concerning the within-day association can be made. Having only one end-of-day assessment limited the capability to address directionality of such within-day association. Future studies should collect multiple assessments within each day to more fully capture this dynamic process. Third, our inability to identify individual-level characteristics that distinguished these two clusters was likely due to small person-level sample size. It is still possible that there can be other characteristics that distinguish these two clusters. Thus, larger sample sizes are very much needed. Fourth, it is important to note that young adults in this sample maintained abstinence without any lapse or relapse during the study period. Therefore, the day-to-day associations found for college students with long-term stable recovery who belong to a CRC can be different from individuals who are currently in treatment or those who have recently left treatment and are fighting to maintain abstinence without the assistance of a supportive community. Thus, it is important to apply similar data collection methods and analytic techniques to investigate intraindividual process and distal relapse outcomes in different recovering populations in a longer time span.

Conclusion

There is substantial interindividual heterogeneity in intraindividual patterns of day-to-day recovery process. Subgroups of recovering individuals could maintain daily abstinence in face of daily relapse risk through distinct mechanisms. These findings highlight the value of applying an idiographic approach to understanding the dynamic daily process of abstinence maintenance among those in sustained recovery. This approach and the corresponding analytic technique have the potential to identify targets for adaptive and tailored intervention to maximize intervention effects and to achieve optimal outcomes, for example, by tailoring quality and dosage of positive social experiences and supports during recovery process.

Footnotes

Authors’ Note

Portions of this article were presented at the Society for Prevention Research annual meeting, Washington, DC, May 2012, and at the Society for Research in Child Development special themed meeting, Transitions from Adolescence to Adulthood, Tampa, FL, October 2012.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The corresponding author was supported by the Prevention and Methodology Training (PAMT) fellowship from the College of Health and Human Development of the Pennsylvania State University.

References

Baker

T. B.

Piper

M. E.

McCarthy

D. E.

Majeskie

M. R.

Fiore

M. C.

(2004). Addiction motivation reformulated: An affective processing model of negative reinforcement. Psychological Review, 111, 33–51.

Bohn

M. J.

Krahn

D. D.

Staehler

B. A.

(1995). Development and validation of an initial measure of drinking urges in abstinent alcoholics. Alcoholism: Clinical and Experimental Research, 19, 600–606.

Cicchetti

Rogosch

F. A.

(1996). Equifinality and multifinality in developmental psychopathology [Editorial]. Development and Psychopathology, 8, 597–600.

Cleveland

H. H.

Harris

K. S.

(2010). The role of coping in moderating within-day associations between negative triggers and substance use cravings: A daily diary investigation. Addictive Behaviors, 35, 60–63.

Collins

L. M.

Murphy

S. A.

Bierman

K. L.

(2004). A conceptual framework for adaptive preventive interventions. Prevention Science, 5, 185–196.

Dimeff

L. A.

Marlatt

G. A.

(1998). Preventing relapse and maintaining change in addictive behaviors. Clinical Psychological: Science and Practice, 5, 513–525.

Dobkin

P. L.

Civita

M. D.

Paraherakis

Gill

(2002). The role of functional social support in treatment retention and outcomes among outpatient adult substance abusers. Addiction, 97, 347–356.

Donovan

D. M.

(1996). Assessment issues and domains in the prediction of relapse. Addiction, 91, 29–36.

Gates

K. M.

Molenaar

P. C. M.

Hillary

F. G.

Ram

Rovine

M. J.

(2010). Automatic search for fMRI connectivity mapping: An alternative to Granger causality testing using formal equivalences among SEM path modeling, VAR, and unified SEM. Neuroimage, 50, 1118–1125.

10.

Hamaker

E. L.

Dolan

C. V.

Molenaar

P. C. M.

(2005). Statistical modeling of the individual: Rationale and application of multivariate stationary time series analysis. Multivariate Behavioral Research, 40, 207–233.

11.

Hoeppner

B. B.

Goodwin

M. S.

Velicer

W. F.

Mooney

M. E.

Hatsukami

D. K.

(2008). Detecting longitudinal patterns of daily smoking following drastic cigarette reduction. Addictive Behavior, 33, 623–639.

12.

Humphreys

Mankowski

Moos

R. H.

Finney

J. W.

(1999). Do enhanced friendship networks and active coping mediated the effect of self-help groups on substance use? Annals of Behavioral Medicine, 21, 54–60.

13.

Humphreys

Moos

R. H.

Cohen

(1997). Social and community resources and long-term recovery from treated and untreated alcoholism. Journal of Studies on Alcohol, 58, 231–238.

14.

Jöreskog

K. G.

Sörbom

(2006). LISREL 8.8 for Windows [Computer software]. Lincolnwood, IL: Scientific Software International.

15.

Kim

Zhu

Chang

Bentler

P. M.

Ernst

(2007). Unified structural equation modeling approach for the analysis of multisubject, multivariate functional MRI data. Human Brain Mapping, 28, 85–93.

16.

Larimer

M. E.

Cronce

J. M.

(2007). Identification, prevention, and treatment revisited: Individual-focused college drinking prevention strategies 1999-2006. Addictive Behavior, 32, 2439–2468.

17.

Love

James

Willner

(1998). A comparison of two alcohol craving questionnaire. Addiction, 93, 1091–1102.

18.

Marlatt

G. A.

Gordon

J. R.

(1985). Relapse prevention: Maintenance strategies in the treatment of addictive behaviors. New York, NY: Guilford Press.

19.

McKay

J. R.

(1999). Studies of factors in relapse to alcohol, drug and nicotine use: A critical review of methodologies and findings. Journal of Studies on Alcohol, 60, 566–576.

20.

McKay

J. R.

(2009). Continuing care research: What we have learned and where we are going. Journal of Substance Abuse Treatment, 36, 131–145.

21.

McKay

J. R.

Franklin

T. R.

Patapis

Lynch

K. G.

(2006). Conceptual, methodological, and analytical issues in the study of relapse. Clinical Psychological Review, 26, 109–127.

22.

McKay

J. R.

Weiss

R. V.

(2001). A review of temporal effects and outcome predictors in substance abuse treatment studies with long-term follow-ups: Preliminary results and methodological issues. Evaluation Review, 25, 113–161.

23.

Molenaar

P. C. M.

(1985). A dynamic factor model for the analysis of multivariate time series. Psychometrika, 50, 181–202.

24.

Molenaar

P. C. M.

(2004). A manifesto on psychology as idiographic science: Bringing the person back into scientific psychology, this time forever. Measurement: Interdisciplinary Research and Perspectives, 2, 201–218.

25.

Molenaar

P. C. M.

Campbell

C. G.

(2009). The new person-specific paradigm in psychology. Current Directions in Psychological Science, 18, 112–117.

26.

Moos

R. H.

Moos

B. S.

(2007). Protective resources and long-term recovery from alcohol use disorders. Drug and Alcohol Dependence, 86, 46–54.

27.

Muthén

Shedden

(1999). Finite mixture modeling with mixture outcomes using the EM algorithm. Biometrics, 55, 463–469.

28.

Ridenour

T. A.

Meyer-Chilenski

Reid

E. E.

(2012). Developmental momentum toward substance dependence: Natural histories and liability of risk factors in youth experiencing chronic stress. Drug and Alcohol Dependence, 123, S87–S98.

29.

Ridenour

T. A.

Pineo

T. Z.

Maldonado

M. M.

Hassmiller

L. K.

(2013). Toward rigorous idiographic research in prevention science: Comparison between three analytic strategies for testing preventive intervention in very small samples. Prevention Science, 14, 267–278.

30.

Robinson

J. D.

Lam

C. Y.

Carter

B. L.

Wetter

D. W.

Cinciripini

P. M.

(2012). Negative reinforcement smoking outcome expectancies are associated with affective response to acute nicotine administration and abstinence. Drug and Alcohol Dependence, 120, 196–201.

31.

Ruehlman

L. S.

Karoly

(1991). With a little flak from my friends: Development and preliminary validation of the test of negative social exchange (TENSE). Journal of Consulting and Clinical Psychology, 3, 97–104.

32.

Scott

C. K.

Foss

M. A.

Dennis

M. L.

(2005). Pathways in the relapse-treatment-recovery cycle over 3 years. Journal of Substance Abuse Treatment, 28, S63–S72.

33.

Shiffman

(2009). Ecological Momentary Assessment (EMA) in studies of substance use. Psychological Assessment, 21, 486–497.

34.

Shiffman

Stone

A. A.

Hufford

M. R.

(2008). Ecological momentary assessment. Annual Review of Clinical Psychology, 4, 1–32.

35.

Shiyko

M. P.

Lanza

S. T.

Tan

Shiffman

(2012). Using the time-varying effect model (TVEM) to examine dynamic associations between negative affect and self confidence on smoking urges: Differences between successful quitters and relapsers. Prevention Science, 13, 288–299.

36.

SPSS Inc. (2001). The SPSS two step cluster component. A scalable component enabling more efficient customer segmentation. White paper—Technical report, Chicago. Retrieved from http://www.spss.ch/upload/1122644952_The%20SPSS%20TwoStep%20Cluster%20Component.pdf

37.

Velicer

W. F.

Colby

S. M.

(2005). A comparison of missing data procedures for ARIMA time series analysis. Educational and Psychological Measurement, 65, 596–615.

38.

Watson

Clark

L. A.

Tellegen

(1988). Development and validation of brief measures of positive and negative affect: The PANAS scale. Journal of Personality and Social Psychology, 24, 1063–1070.

39.

Weisner

Matzger

Kaskutas

L. A.

(2003). How important is treatment? One-year outcomes of treated and untreated alcohol-dependent individuals. Addiction, 98, 901–911.

40.

Witkiewitz

Bowen

(2010). Depression, craving, and substance use following a randomized trial of mindfulness-based relapse prevention. Journal of Consulting and Clinical Psychology, 78, 362–374.

41.

Witkiewitz

van der Maas

H. L. J.

Hufford

M. R.

Marlatt

G. A.

(2007). Nonnormality and divergence in posttreatment alcohol use: Reexamining the project MATCH data “another way.” Journal of Abnormal Psychology, 116, 378–394.

42.

Witkiewitz

Marlatt

G. A.

(2004). Relapse prevention for alcohol and drug problems: That was Zen, this is Tao. American Psychologist, 59, 224–235.

43.

Zheng

Wiebe

R. P.

Cleveland

H. H.

Molenaar

P. C. M.

Harris

K. S.

(2013). An idiographic examination of day-to-day patterns of substance use craving, negative affect and tobacco use among young adults in recovery. Multivariate Behavioral Research, 48, 241–266.