Hedeker & Mermelstein 1 Mood Changes Associated with Smoking in Adolescents: An Application of a Mixed- Effects Location Scale Model for Longitudinal Ecological Momentary Assessment (EMA) Data

Mixed models are often used for analysis of Ecological Momentary Assessment (EMA) data. The error variance, which indicates the degree of within-subjects variation, is usually considered to be homogeneous across subjects. Modeling this variance can shed light on interesting hypotheses in psychological research. We describe how the within-subject variance can be modeled in terms of covariates to examine the stability/lability of mood within subjects, focusing on positive and negative mood and the degree to which these moods change as a function of smoking in adolescents. The data are drawn from an EMA study of adolescent smoking. Participants were 130 adolescents, either in 9th or 10th grade, who provided EMA mood reports following smoking events at two or more measurement waves. We focused on two mood outcomes: changes in the subject’s negative and positive affect, after minus before smoking, and several covariates: gender, wave, and smoking level. The results indicate that the variation in smoking-related mood changes were diminished as a person increased their smoking level. Thus, our analyses revealed an increased consistency of smoking-related mood changes as smoking level increased within a subject. Hedeker & Mermelstein 3 Introduction Modern data collection procedures, such as ecological momentary assessments (EMA; Smyth & Stone, 2003; Stone & Shiffman, 1994), experience sampling (Feldman Barrett & Barrett, 2001; Scollon, Kim-Prieto, & Diener, 2003), and diary methods (Bolger, Davis, & Rafaeli, 2001) yield relatively large numbers of subjects and observations per subject, and data from such designs are sometimes referred to as intensive longitudinal data (Walls & Schafer, 2006). Analysis of EMA data using mixed models (also known as multilevel or hierarchical linear models) is well-described by Schwartz and Stone (2007). Additionally, Moghaddam and Ferguson (2007) analyzed EMA data using mixed models to examine smoking-related changes in mood. These articles focus on the effects of covariates, either subject-varying or time-varying, on the EMA mean responses. Here we extend this approach by examining the degree to which covariates influence the within-subjects variation inherent in the EMA data. A few articles have described approaches for examining determinants of betweenand within-subjects variance from EMA studies. Penner, Shiffman, Paty, and Fritzsche (1994) used basic descriptive statistical methods to examine relations among within-subject variation in several mood variables. Jahng, Wood, and Trull (2008) described generalized mixed models to analyze within-subject differences in sequential EMA mood responses, specifically characterizing these as mean square successive differences. Hedeker, Mermelstein, Berbaum, and Campbell (2009) described a mixed model that included determinants of the betweensubjects variance, while Hedeker, Mermelstein, and Demirtas (2008) developed a model that additionally allowed determinants of the within-subjects variance plus a random subject scale effect. This model is referred to as a mixed-effects location scale model because subjects have both random location and scale effects. Models with random scale effects have been described in several articles where interest centers on variance modeling and/or accounting Hedeker & Mermelstein 4 for heterogeneous variation across individuals or clusters (Chinchilli, Esinhart, & Miller, 1995; Cleveland, Denby, & Liu, 2002; James, James, Venables, Dry, & Wiskich, 1994; Leckie, 2010; Lin, Raz, & Harlow, 1997; Myles, Price, Hunter, Day, & Duffy, 2003). In this chapter, we extend the mixed-effects location scale model to focus on the variation of mood change that is associated with smoking across measurement waves, and the degree to which subject and wave characteristics influence the variation in mood changes. Also, while Hedeker et al. (2008) only considered random subject intercepts for the one wave of EMA data, here we allow random subject time trends for the multiple waves of EMA data. We further consider a three-level model that treats observations nested within waves within subjects. To aid in making this class of models accessible, sample computer syntax is provided in the Appendix. Adolescent smoking, mood, and variability Many prominent models of cigarette smoking maintain that smoking is reinforcing, and that smoking can relieve negative affect (Kassel, Stround, & Paronis, 2003; Khantzian, 1997). Indeed, both adults and adolescents often claim that smoking is relaxing and reduces emotional distress (Chassin, Presson, Rose, & Sherman, 2007; Kassel & Hankin, 2006). However, although the relation between mood and smoking has received substantial empirical attention for adult smokers, much less is known about the acute changes in mood with smoking among adolescents. The present study, with its focus on real-time assessments of mood and smoking among adolescents, helps to shed light on this important topic. Although there is substantial consensus among both smokers and researchers that smoking helps to regulate affect, most of the empirical work investigating the smoking-mood relation has focused on the examination of changes in mean levels of mood with smoking. Surprisingly, although affect regulation inherently implies the modulation of variability in Hedeker & Mermelstein 5 mood as well, the examination of variability in mood and smoking has largely been neglected. As Hertzog and Nesselroade (2003) noted, describing mean levels of variables is not always adequate for examining key features of developmental change. Variation also conveys important information about the phenomenon of interest. In the case of adolescent smoking and the development of dependence, variation in mood and mood changes may help to explain more of the development of tolerance. Examining individual variability may enhance our ability to predict changes in smoking behavior above and beyond what can be achieved by examining mean information alone. Important, too, in the examination of mood and smoking, is the distinction between within-person and between-person variability. Kassel and colleagues (Kassel & Hankin, 2006; Kassel et al., 2003) have argued persuasively for the need to differentiate causal within-person mechanisms from between-person data. Whether smoking relieves negative affect is essentially a within-person question, and thus analytic models need to similarly differentiate between within-subject and between-subject effects. Much of the research on mood and smoking has also been limited to assessments of negative affect, while ignoring positive affect. This neglect is particularly problematic given the theoretical importance of differentiating between negative reinforcement models of smoking and positive reinforcement models, especially in the development of dependence among adolescents (Tiffany, Conklin, Shiffman, & Clayton, 2004). There is also considerable evidence to support the notion that positive and negative affect are distinct constructs, and not just opposite ends of a continuum (Watson & Tellegen, 1985; Watson, Wiese, Vaidya, & Tellegen, 1999). Thus, in the current study, we assessed both positive and negative affect. Finally, there may well be individual differences in the extent to which adolescents’ moods vary and whether these moods vary with smoking. Identifying potential moderator variables may also help in the prediction of smoking escalation among relatively novice Hedeker & Mermelstein 6 smokers. Indeed, in a previous paper (Hedeker et al., 2009) it was observed that adolescent smoking level was associated with variation in mood changes associated with smoking, diminishing this variance for both positive and negative affect. While this finding was noteworthy, it represents a between-subjects effect of smoking level, rather than addressing the point of whether variation in mood changes associated with smoking diminishes as a person increases their smoking level (a within-subjects effect). Here, we aim to assess this within-subjects effect by modeling the EMA data across several measurement waves as a subject changes their smoking level. We hypothesized that the between-subjects effect of smoking level, that we previously reported, would also be observed as a within-subjects effect. Namely, as adolescents increase their level of smoking across time, their variation in mood changes associated with smoking would diminish. Thus, following along the lines of the development of tolerance with dependence, we hypothesized that as smoking level or experience increased, mood responses to smoking would decrease, as would variability in overall mood. Methods Subjects The data are drawn from a natural history study of adolescent smoking. Participants included in this study were in 9th or 10th grade at baseline, 55.1% female, and reported on a screening questionnaire 6-8 weeks prior to baseline that they had smoked at least one cigarette in their lifetimes. The majority (57.6%) had smoked at least one cigarette in the past month at baseline. Written parental consent and student assent were required for participation. A total of 461 students completed the baseline measurement wave. Of these, 57% were white, 20% Hispanic, 16% black, and 7% of other race. The study utilized a multi-method approach to assess adolescents including self-report questionnaires, a week-long time/event sampling method via hand-held computers (EMA), Hedeker & Mermelstein 7 and detailed surveys. Adolescents carried the hand-held computers with them at all times during a data collection period of seven consecutive days and were trained both to respond to random prompts from the computers and to event record (initiate a data collection interview) in conjunction with smoking episodes. Random prompts and the self-initiated smoking records were mutually exclusive; no smoking occurred during random prompts. Questions concerned place, activity, companionship, mood, and other subjective variables. The handheld computers dated and time-stamped each entry. Following the baseline assessment, subjects completed additional EMA sessions at 6-, 15-, and 24-month follow-ups, for a total of four EMA measurement waves. Subject retention was good, with 405, 360, and 385 subjects completing the EMA sessions at these three follow-ups, respectively. Since estimation of model parameters is based on a full-likelihood approach, the missing data are assumed to be “ignorable” conditional on both the model covariates and the observed responses of the dependent variable (Laird, 1988). In longitudinal studies, ignorable nonresponse falls under the “missing at random” (MAR) mechanism of Rubin (1976), in which the missingness depends only on observed data. As Molenberghs et al. (2004) indicate, MAR is a relatively weak assumption, especially as compared to the more stringent missing completely at random (MCAR) assumption, and one that we will make here. For the interested reader, extended not missing at random (NMAR) approaches are described in Chapter 14 of Hedeker and Gibbons (2006). Because of our interest in comparing mood within subjects from smoking events across measurement waves, we restricted our analysis to subjects who provided two or more waves of data; where at each wave the subject provided at least two smoking events. In all, there were 130 such subjects with data from a total of 3,388 smoking events. Of these, 47, 39, and 44 subjects provided data at two, three, and four waves, respectively. The number of subjects at each measurement wave equaled 116 (baseline), 91 (6 months), 92 (15 months), and 88 (24 Hedeker & Mermelstein 8 months), and the average number of smoking events equaled 7.14 (range = 2 to 42), 7.65 (range = 2 to 32), 9.97 (range = 2 to 43), and 10.76 (range = 2 to 49) at these same four waves, respectively. Measures Negative and Positive Affect Two mood outcomes were considered: measures of the subject’s negative and positive affect (denoted NA and PA, respectively) at smoking episode. Both of these measures consisted of the average of several individual mood items, each rated from 1 to 10 with “10” representing very high levels of the attribute, that were identified via factor analysis. Specifically, PA consisted of the following items that reflected subjects’ assessments of their positive mood: I felt happy, I felt relaxed, I felt cheerful, I felt confident, and I felt accepted by others. Similarly, NA consisted of the following items: I felt sad, I felt stressed, I felt angry, I felt frustrated, and I felt irritable. For the smoking events, participants rated their mood “before” and “now after smoking.” Considering the five items of the “before” (and “now after smoking” ) PA mood assessments, Cronbach’s alpha was equal to .84 (.77), .81 (.78), .85 (.83), and .83 (.82) at baseline, 6-, 9-, and 24-months, respectively. Similarly, in terms of the NA mood assessments, Cronbach’s alpha equaled .90 (.90), .92 (.91), .88 (.91), and .93 (.90) at baseline, 6-, 9-, and 24-months, respectively Because of our interest in smoking-related mood change, we used the difference (now-before) as our measure of reported mood change associated with smoking. Gender and Wave As covariates, we considered gender and measurement wave with the variables Male (coded 0=female or 1=male) and Wave (coded 0=baseline, 1=6 months, 2.5=15 months, and 4=24 months). In our selected sample of 130 subjects, 46% were males. Hedeker & Mermelstein 9 Smoking Level As a time-varying (within-subjects) measure of a subject’s smoking level, we used the number of smoking events that a subject reported at a given measurement wave (denoted as NumSmk). To separate the betweenand within-subjects effects of this time-varying variable on mood change, as described in Begg and Parides (2003), we also included the subject’s mean of NumSmk as a covariate (denoted as AvgSmk). By including both the wave-varying NumSmk and the subject-varying AvgSmk, we can estimate, respectively, both the withinsubjects and between-subjects effects of smoking level on mood change. The between-subject effect represents the association of a person’s average smoking level with their average change in mood (both averages being taken over time). Conversely, the within-subjects effect indicates how a person’s mood change differs as their level of smoking varies over waves. The latter is of primary interest here as it represents the degree to which a person’s mood response to smoking (now – before) changes as their smoking level varies across time. Finally, because of the relatively large numerical range of these variables, to ease computation and interpretation, we divided both by a factor of 10 so that the coefficients of these variables represent changes attributable to 10 smoking events (rather than a single smoking event). Also, to increase the interpretation of the intercept-related parameters we centered these two smoking level variables at the value of 10 smoking reports. Data Analysis Consider the following mixed-effects regression model for the measurement y , either smoking-related change in NA or PA, of subject i ( 1 2     i ... N subjects) on occasion j ( 1 2     i j ... n occasions): 0 0 1 1 2 3 4 ( ) ( ) ij i ij ij i i ij i y                 AvgSmk Wave Male NumSmk (1) Hedeker & Mermelstein 10 Here, the occasions refer to the multiple smoking events that a subject provides, which, based on our inclusion criteria, are obtained at two or more measurement waves for each subject. The random subject effect 0i  indicates the influence of individual i on his/her mood change at baseline, while 1i  represents how a subject’s mood change varies over time. Both of these reflect individual deviations relative to the population intercept and slope, 0  and 1  . The inclusion of the random slope 1i  is important here because the data are collected across multiple waves. With two random subject-specific effects, the population distribution of intercept and slope deviations is assumed to be a bivariate normal (0 ) N  Σ , where  Σ is the 2 2  variance-covariance matrix given as: