Smoking patterns and dependence : Contrasting chippers and heavy smokers

Acknowledgement: Jean Paty is now at invivodata, inc., Pittsburgh, Pennsylvania.
This research was supported by Grant DA04281 from the National Institute on Drug Abuse. The research served as a master's thesis in psychology for Jean Paty, under the direction of the Saul Shiffman. Saul Shiffman and Jean Paty are cofounders and shareholders of invivodata, inc., which provides electronic diaries for clinical research.
We are grateful for comments on the statistical analysis from Linda Collins and Joseph Schwartz and for general comments from Michael Sayette, the assistance of Maryann Gnys and Jon Kassel in conducting the study, the assistance of Stuart Ferguson and Qianyu Dang in data analysis, and the editorial assistance of Yolanda DiBucci.

Nicotine dependence is typically the driving motive behind tobacco use and cigarette smoking, and most current smokers are dependent (U.S. Department of Health and Human Services, 1988). However, studies have documented that some people are able to engage in stable patterns of light smoking without progressing to dependence (Hajek, West, & Wilson, 1995; Shiffman, 1989; Shiffman, Paty, Kassel, Gnys, & Zettler-Segal, 1994). Most people who smoke regularly progress to dependence (Shiffman & Paton, 1999); intermittent smoking is generally an unstable pattern leading either to progression to regular smoking or to desistence (e.g., Zhu, Sun, Hawkins, Pierce, & Cunningham, 2003). However, we have demonstrated that some intermittent smokers can demonstrate substantial resistance to dependence, sustaining a regular pattern of light smoking for years without developing dependence (Shiffman, Paty, et al., 1994). These smokers—dubbed “chippers” (CHs)—absorb nicotine from cigarettes in normal amounts and metabolize nicotine normally (Brauer, Hatsukami, Hanson, & Shiffman, 1996; Shiffman, Fischer, Zettler-Segal, & Benowitz, 1990; Shiffman et al., 1992) but, despite having smoked tens of thousands of cigarettes, show few signs of nicotine dependence (Shiffman, 1989; Shiffman, Paty, et al., 1994) and no sign of withdrawal even after several days of abstinence (Shiffman, Paty, Gnys, Kassel, & Elash, 1995). CHs do not seem to smoke to avoid nicotine depletion or to stave off withdrawal. These findings suggest that repeated exposure to smoking may not in itself inevitably lead to nicotine dependence. Low-rate smoking and less-than-daily smoking are common among teen smokers early in their smoking careers, but this pattern is typically subsequently displaced by daily and higher-rate smoking as smoking progresses (Gilpin et al., 2001). However, CHs seem to have failed to progress or “mature” into heavier and more consistent smoking; we have hypothesized that they may display patterns of smoking associated with earlier stages of the smoking career (Shiffman, 1991; Shiffman, Paty, et al., 1994).

Smoking patterns can provide some clues to smoking motives. For example, if smokers smoke when stressed, this would bolster the notion that they are smoking to mitigate negative affect (Shiffman et al., 2002; Shiffman, Kassel, Paty, Gnys, & Zettler-Segal, 1994). In recent analyses of heavy smokers' (HSs') smoking patterns, we found that their ad lib smoking was not substantially associated with affective state (Shiffman et al., 2002; Shiffman, Paty, Gwaltney, & Dang, 2004). This lack of association between smoking and situational stimuli was explained in terms of HSs' addictive drive to smoke regularly to maintain blood nicotine levels and avoid lapsing into nicotine withdrawal (Shiffman et al., 2002), thus overriding any association between smoking and other stimuli. Contrasting HSs' smoking patterns with those of CHs may help test this hypothesis. If this explanation of HSs' smoking is valid, then we would expect CHs' smoking to demonstrate much stronger association with situational stimuli (which we refer to as “stimulus control”).

Several studies (Chassin, Presson, & Sherman, 1984; Hajek et al., 1995; Jarvik, Killen, Varady, & Fortman, 1993; Shiffman, Kassel, et al., 1994) have used global retrospective self-report questionnaires (“smoking typology scales”) to ask CHs about their smoking patterns and motives. Compared with HSs, CHs tend to emphasize social motives, those related to the sensory pleasure of smoking and those associated with pleasant relaxation or smoking after meals (as opposed to motives associated with craving, relief of negative affect, stimulation, and habitual or automatic smoking). Thus, CHs may be what M. A. H. Russell, Peto, and Patel (1974) called “indulgent” smokers. However, questionnaires assessing smoking patterns are fraught with psychometric problems and do not accurately reflect smoking patterns (Shiffman, 1993; Shiffman & Prange, 1988; Tate, Schmitz, & Stanton, 1991). CHs may also fit the pattern described by Moran, Wechsler, and Rigotti, (2004) as “social smoking,” defined as smoking primarily with others and primarily as a social activity, implying smoking driven by external social motives rather than by internal or pharmacological motives.

In this study, we used data on smoking and nonsmoking situations collected in real time in real-world settings with ecological momentary assessment (EMA; Shiffman, 1993; Shiffman et al., 2002; Shiffman, Paty, et al., 2004; Stone & Shiffman, 1994) to analyze CHs' smoking patterns. In an extreme-groups design, we contrasted CHs' patterns with those of closely matched HSs. We examined the hypothesis that CHs would be particularly likely to smoke when with others (Moran et al., 2004) and when drinking alcohol (Istvan & Matarazzo, 1984; Shiffman & Balabanis, 1995), a pattern resembling the behavior of smokers early in their smoking careers (McKee, Hinson, Roundsaville, & Petrelli, 2004).

We also assessed CHs' urges to smoke, both when smoking and when not smoking. Whereas some studies have suggested that CHs experience little craving (Davies, Willner, & Morgan, 2000; Sayette, Wilson, & Elias, 1992; Shiffman et al., 1995), others have suggested that CHs do experience craving, especially when exposed to smoking cues (Davies et al., 2000; Sayette et al., 1992, 2003), suggesting that CHs' craving might vary considerably according to the circumstances. We expected that CHs might experience craving in the occasional situation when they smoke but not when not smoking, whereas HSs may report craving even when not smoking (indicating their dependence).

We focused on stimulus control of smoking as a key differentiator of chipping versus heavy smoking. As dependence develops, smoking is expected to become increasingly driven by the ebb and flow of blood nicotine levels. In particular, because nicotine is cleared quickly from the body and thus has to be “topped off” frequently, for dependent smokers, the need to regularly replenish nicotine would tend to override and erode the stimulus control that might otherwise be exercised by situational stimuli. Whereas even dependent smokers are expected to have some discretion in when they smoke (Kozlowski & Herman, 1984), the diminution of stimulus control and the development of stereotypy may be a hallmark of dependence, as use of the drug shifts from being a purely discretionary pleasure that accompanies certain activities, but which the smoker can take or leave, to a compulsive pattern of regular and stereotypic self-administration (Edwards, 1986; Shiffman, Waters, & Hickcox, 2004). Thus, we hypothesized that HSs' smoking would demonstrate less stimulus control than would CHs' smoking.

Whereas an analysis of particular antecedents can examine the association of smoking with particular situational variables, an analysis of overall stimulus control must look past associations with particular variables and the direction of those associations. For example, some CHs may tend to smoke when working and stressed, whereas others smoke when eating and relaxed. Both patterns demonstrate stimulus control over smoking, but a groupwise nomothetic analysis might show no effect. To assess the degree of stimulus control, we performed idiographic analyses, in which we modeled stimulus associations separately for each smoker and statistically estimated the overall magnitude of the associations across a range of situational stimuli. We then entered these estimates into a second-level analysis, which contrasted the average levels of stimulus control among CHs and HSs.

In our assessment, we avoided relying on global assessments or recall of smoking patterns (Hammersley, 1994; Shiffman, 1993); instead, we used EMA (Stone & Shiffman, 1994) methods to collect real-time data about smoking and nonsmoking episodes in subjects' natural environments. Subjects noted (with a palmtop computer that time tagged records to document timely entry; see Stone, Shiffman, Schwartz, Broderick, & Hufford, 2002) when they were smoking and completed a situational and affective assessment. As a control or contrast for smoking occasions (Paty, Kassel, & Shiffman, 1992; Shiffman, 2004), the computer beeped subjects for a similar assessment at randomly selected nonsmoking times (Csikszentmihalyi & Larson, 1987; Shiffman et al., 2002).

This was an extreme-groups case-control design, contrasting smoking patterns in a group of CHs selected to demonstrate resistance to nicotine dependence with a matched group of relatively HSs.

Participants were 26 CHs and 28 HSs recruited through the media (Shiffman, Paty, et al., 1994). CHs were highly selected to demonstrate a pattern of regular smoking without progression. They had to average no more than 5 cigarettes per day while smoking at least 4 days per week. HSs smoked 20 to 40 cigarettes daily. The inclusion criteria were confirmed by collateral reports and by biochemical assays where possible (Shiffman, Paty, et al., 1994). HSs were recruited to closely match CHs on gender, age, number of years of smoking (and, consequently, on age at initiation), and the nominal nicotine delivery of their cigarettes (Federal Trade Commission method).

Shiffman, Paty, et al. (1994) described and justified the selection criteria in detail. The emphasis was on identifying smokers whose smoking was stable rather than transitional. Subjects were included only if they were not planning to quit or cut back smoking, were over 23 years old, were Caucasian, did not use tobacco products other than cigarettes, and had smoked for at least 4 years with no substantial changes in the past 2 years, no more than three quit efforts in the past 2 years, and no cessation of smoking for 2 days or more in the past 6 months. Table 1 shows the demographics and smoking history of CHs and HSs.

The methods of the study have been described in Shiffman, Paty, et al. (2004) and Shiffman et al. (2002). The study was conducted in 1988 in Pittsburgh, PA, at a time when few regulations restricting smoking were in place. Participants were trained to use a handheld computer designed to allow data collection in near real time. The electronic diary (ED; see Shiffman et al., 1995) was a PSION Organizer II palmtop (5.6″ × 3.1′ × 1.1″; 8.8 oz; PSION, Ltd., London, England) with a 2-line, 16-character LCD screen. For 2 weeks, participants monitored their ad libitum smoking throughout the waking day by recording each cigarette on the ED immediately before smoking. All cigarettes recorded by CHs were assessed. Because HSs smoked often, ED was programmed not to administer assessments for all their cigarettes but instead to select smoking occasions at random, such that approximately 5 cigarettes per day were assessed (smoking). The appropriate sampling ratio was based on subjects' self-reported smoking frequency; for example, for a subject who reported smoking 20 cigarettes per day, a random 25% of cigarette entries were selected for assessment. Even when no assessment was administered, the ED made a record of each cigarette. In addition, participants were beeped by the ED four to five times daily at random times to complete a similar assessment while they were not smoking (nonsmoking).

Identical assessments of smoking and nonsmoking situations were completed on the screen, one item at a time. Participants rated mood and urge to smoke on a 4-point scale ranging from 1 (NO!!) to 4 (YES!!). Mood ratings were made for adjectives (listed below) derived from the circumplex model of affect (J. Russell, 1980) as well as bipolar items on overall affect and arousal. We used exploratory factor analysis and multidimensional scaling to score the affect items into three scales: Negative Affect (high-loading items, in descending order of factor loadings, with italics indicating reverse scoring: happy, calm, tense, delighted, frustrated, overall affect; Cronbach's alpha = .80), Arousal (arousal, tired, overall arousal; α = .75), and Low Mood (sad, miserable, bored, frustrated; α = .75). We used factor scores (standardized scores, with M = 0 and SD = 1.0 over the whole sample) derived from varimax rotation because they yield uncorrelated scale scores. The Negative Affect and Arousal factors were bipolar, with low scores representing positive affect and low arousal, respectively. Participants also noted current activities and location (see Table 2).

Participants contributed data from a total of 831 days (CHs: 395; HSs: 436), encompassing 8,496 assessments (CHs: 3,889; HSs: 4,597)—4,971 nonsmoking assessments (CHs: 2,591; HSs: 2,380) and 3,525 smoking assessments (CHs: 1,308; HSs: 2,217). Individual smoking and nonsmoking observations were analyzed and were not aggregated (see Shiffman et al., 2002). We used generalized estimating equations (GEE) to account for the nesting of multiple observations within subjects and the autocorrelations among them, specifying an AR(1) autocorrelation structure (Zeger, Liang, & Albert, 1988). We have previously presented an analysis of HSs' smoking patterns (Shiffman, Paty, et al., 2004). The analysis of CHs' data, essentially a logistic regression, treated observation type (smoking vs. nonsmoking) as a categorical dependent variable modeled as a function of a situational variable (mood, activity, situation). We report odds ratios, confidence intervals, and statistical significance. Given the number of tests, we emphasize relationships where p < .01. We used point-biserial or phi correlation coefficients to calculate for each participant the degree of association between smoking and its antecedents, and we report the average within-subject correlation coefficient as another effect-size estimate. Comparison of situational associations between CHs and HSs also used GEE analysis but extended the models to include a Group effect and a Group × Antecedent interaction. The interaction, which was the focus of the analysis, essentially tests the ratio of the two odds ratios seen among CHs and those seen among HSs.

To assess the degree to which smoking was under stimulus control of various stimuli, by subject, we performed a two-level analysis, starting with an idiographic analysis with each subject (Bryk & Raudenbush, 1992). For each subject, we constructed a logistic regression equation predicting smoking (vs. nonsmoking) from a set of situational antecedents. Separate analyses were conducted for blocks of variables, as defined in Table 2: activity, eating and drinking, location, social setting, mood, and time. (Models combining all variables could not be estimated because of collinearity.) For each subject's equation, we estimated the area under the curve (AUC) for the receiver operating characteristic (ROC) curve characterizing the degree of prediction of smoking from the independent variables. The magnitude of AUC-ROC reflects the degree to which smoking is predictable, independent of which particular variables (antecedents) were predictive. To compare the degree to which smoking was predictable from situational variables for CHs and HSs, we entered the individually computed AUC-ROC values into a second-level, between-groups analysis (Bryk & Raudenbush, 1992), using t tests and nonparametric median tests. To account for the fact that some AUC-ROC values may have been more reliably estimated than others (on the basis of the number of observations and the estimated AUC-ROC value itself), we weighted the cases by the inverse of the standard error of AUC-ROC, which was computed with the approach of Hanley & McNeil (1982). Weighted and unweighted analyses yielded identical results; we report the weighted values.

An important virtue of the AUC-ROC as an indicator of the magnitude of the relationship is that AUC-ROC has a conceptually meaningful interpretation. AUC-ROC is interpretable as the probability of correctly identifying a smoking observation from a randomly chosen pair of smoking and nonsmoking observations, given the predictor variables (Hanley & McNeil, 1982). That is, if AUC-ROC is 0.75, this means that in all possible pairwise comparisons of a smoking and nonsmoking observation, the model would enable one to correctly identify the smoking observation 75% of the time. AUC-ROC ranges from 0.5 (chance identification) to 1.0 (perfect identification).

Table 1 shows the characteristics of the samples. The groups closely resemble each other on many variables because of matching and differ systematically in smoking behavior and history because of selection as CHs and HSs. Whereas CHs smoked an average of 3.9 (SD = 1.2) cigarettes per day, 6 days a week, the HSs smoked an average of 31.4 (SD = 6.6) cigarettes daily.

Subjects provided data for an average of 15.4 days. On average, CHs contributed 100 random assessments and 50 smoking assessments. HSs contributed 85 random assessments and 79 smoking assessments; in addition, each HS recorded an average of 275 cigarettes that were not assessed.

Compliance with protocol instructions was first assessed by self-report, with questions phrased to encourage admission of noncompliance. Despite the permissive set, the subjects reported carrying the ED 94% of the time and recording 97% of cigarettes smoked. More objectively, half the subjects responded to at least 89% of all random prompts within the 2 min allotted; only 1 out of 4 subjects missed even one prompt per day. Compliance with recording of cigarettes was assessed by correlating the number recorded to exhaled levels of carbon monoxide (CO), a biochemical marker of recent smoking (Vogt, Selvin, Widdowson, & Hulley, 1977). Afternoon CO levels correlated with the number of cigarettes recorded in the preceding 4 hr among both CHs (r = .56, p < .01) and HSs (r = .52, p < .01). The number of cigarettes recorded did not change significantly between the first 2 and last 2 days of the study for CHs (3.3 vs. 3.7; ns) but did drop modestly for HSs (28.7 vs. 24.6; p < .005), at an average rate of 0.33 cigarettes per day. CO levels showed no change (CHs: Day 2 = 7.6 ppm [SD = 2.9], last day = 7.25 ppm [SD = 3.0], ns; HSs: Day 2 = 27.6 ppm [SD = 11.8], last day = 27.3 ppm [SD = 7.5], ns).

Table 2 summarizes the data obtained from CHs when they were smoking and when sampled at random when not smoking, comparing the antecedents of those two contexts. The analysis evaluates the change in the odds of smoking with changes in the situational antecedent, expressed as an odds ratio. In other words, it evaluates whether the probability of smoking (vs. not smoking) differed in a particular circumstance (e.g., drinking alcohol) compared with its complement (not drinking alcohol). CHs were more likely to smoke in certain locations; the odds of smoking increased when CHs were at home or at a bar or restaurant (compared with all other locations). They were less likely to smoke in situations they characterized as “other” (including outdoors). Being at work, at others' homes, and in a car did not significantly affect the probability of smoking.

CHs' smoking also was associated with concurrent activities. The probability of smoking increased when they were relaxing, socializing, or doing nothing and decreased when they were working or when they were engaged in activities classified as “other.” CHs were also more likely to smoke when eating, drinking coffee, or drinking alcohol. The relationship with alcohol was particularly strong: Drinking alcohol increased the odds of smoking nearly fivefold.

Collectively, the activities associated with CHs' smoking suggested that CHs were more likely to smoke in those activities that M. A. H. Russell et al. (1974) described as “indulgent.” To capture this, we coded situations as indulgent if they involved eating, drinking alcohol, relaxing, socializing, or doing nothing and did not involve working. Such situations accounted for 75% of CHs' cigarette smoking; the odds of smoking were more than 4 times greater in indulgent settings than in others (see Table 2).

CHs' smoking also seemed to be influenced by others' smoking. They were more likely to smoke when others were smoking and less likely to smoke when others were not smoking. However, these data do not support the hypothetical characterization of CHs as social smokers. CHs smoked almost half their cigarettes when they were alone and were actually more likely to be alone when they were smoking than when they were not smoking.

The probability of CHs' smoking was not influenced by their affective state. There was no linear association between smoking and negative affect. We considered that CHs might smoke under both negative and positive affect, which would “wash out” in a linear analysis of a bipolar affect scale. Accordingly, we evaluated the quadratic trend in affect, which would capture a tendency to smoke when either feeling good or feeling bad. No significant association was found. One minor finding was that a single item rating current arousal indicated that smoking was more likely under low arousal. However, this effect appears to be related to CHs' tendency to smoke later in the day and disappeared when we restricted the analysis to the evening hours. (We also tested individual affect items and the negative affect by arousal interaction; these were unrelated to smoking.)

CHs did report urges to smoke; 16% of their ratings indicated an intensity of 4, the highest value on the scale. These were strongly associated with smoking: 89% of these high-urge occasions were recorded when CHs were smoking. The odds of smoking increased more than eightfold for every 1-point increase in urge intensity. Examination of the probability of smoking at each urge level shows that the likelihood of smoking rises steeply as one crosses the midpoint of the urge scale. On average, CHs reported moderate to strong urges when smoking (3.3 on a 4-point scale), but they denied urges (the average of 1.5 lies between the response options labeled “no??” and those labeled “NO!!”) when not smoking.

Table 2 also shows the data associated with HSs' smoking (from Shiffman, Paty, et al., 2004). A set of analyses evaluated the interaction of group differences with smoking versus nonsmoking differences. A significant interaction indicates significant differences between CHs and HSs in the odds ratios for antecedents. In other words, these analyses ask whether the association between an antecedent (e.g., alcohol) and smoking (vs. nonsmoking) differs for CHs and HSs.

There were several significant differences, some of which are graphed in Figure 1. The activity of relaxing had a stronger association with CHs' smoking than with HSs' smoking. Also, whereas CHs were less likely to be smoking when engaged in “other” activities, the trend among HSs was reversed. There was also a trend for CHs' smoking, compared with HSs' smoking, to be more suppressed when working.

Eating or drinking (food, coffee, and/or alcohol) was associated with a greater increase in the probability of smoking among CHs than among HSs. The differences in response to alcohol were the most striking: CHs showed a much stronger association between drinking and smoking than did HSs. However, the association between smoking and eating was stronger for CHs than for HSs, even when alcohol situations were excluded, suggesting that CHs' smoking was also more associated with other types of consumption. With the omnibus coding of indulgent situations, we found that CHs' smoking was much more associated with indulgent activities than was HSs' ( Figure 2, left panel) smoking.

CHs were less likely to smoke when engaged in “other” activities and in “other” locations (which included outdoors), whereas HSs showed no such relationships. CHs were also less likely to smoke in vehicles, which was positively associated with smoking among HSs. Compared with HSs, CHs were more responsive to others' smoking. The proportion of cigarettes they smoked when others were smoking was similar, but CHs generally spent less time where others were smoking, so their smoking was proportionately more influenced by others' smoking (see Figure 1). There were no group differences in the proportion of cigarettes smoked while alone or in the likelihood of smoking alone.

There were no group differences in the association between affect and smoking, whether for negative affect, arousal, or depressed affect. (We also tested individual affect items and the negative affect by arousal interaction; these showed no interactions between group and smoking.) However, there were significant differences in the relationship between urge and smoking. When smoking, CHs and HSs gave very similar urge ratings, with the majority of ratings indicating the presence of urges ( Figure 3, right panel). However, differences emerged in nonsmoking situations (left panel). Whereas HSs' ratings were generally in the neutral range, with 53% indicating some degree of urge even when not smoking, in these nonsmoking situations, CHs indicated very low urges or even an aversion to smoking (only 15% indicated some degree of urge).

Figure 4 shows the distribution of smoking times for CHs and HSs. It is evident that CHs' smoking is skewed toward later in the day, with most occurring after 5 p.m. (This was not universally true, however: A quarter of CHs smoked the majority of their cigarettes before 5 p.m. Altogether, two thirds of CHs' cigarettes were smoked after 5 p.m., compared with 43% of HSs' cigarettes.) To control for this difference in time of day, we repeated the between-groups analysis with only data collected after 5 p.m. The results were largely identical, with a few exceptions. Two Group × Smoking interactions were no longer significant in the time-controlled analysis: The tendency of CHs' smoking to be more promoted than that of HSs by others' smoking was eliminated and became nonsignificant (from OR = 1.57 to OR = 1.07). Also, the difference in overall arousal went from OR = 0.78 to OR = 0.86, indicating that the arousal effects were due to CHs not smoking earlier in the day. Conversely, two previously nonsignificant interaction effects became significant in the after-5-p.m. data: CHs' smoking was more closely associated with being in bars or restaurants than was HSs' smoking (OR went from 1.66 [all day] to 2.29 [after 5]), and CHs' smoking was also significantly more closely associated with coffee drinking than was HSs' smoking (OR = 1.81 to 2.61). Otherwise, restricting the analysis to observations after 5 p.m. had little directional effect: The ORs varied slightly, of course, but the between-groups differences were equally likely to strengthen or weaken, and there was little change in the average OR. In Figure 2, we illustrate the relative stability of the results with the relationship between smoking and indulgence, as the indulgence variable appears to capture many of the group differences in smoking. Whereas the indulgent settings occur more often in the evening than they do earlier in the day (illustrated by the greater height of the points in the right panel), the differential impact of indulgence on CHs' smoking, compared with that of HSs, is almost unchanged when we limit observations to those after 5 p.m. (The OR went from 3.06 for all day to 2.86 for after 5 p.m.)

We observed that much of CHs' smoking took place during indulgent activities; such situations accounted for 75% of CHs' cigarette smoking but only 55% of HSs' smoking. We speculated that, perhaps, smoking cigarettes in these indulgent situations might be motivated by these situational cues, whereas smoking outside of these situations might be influenced by negative affect. Accordingly, we conducted a GEE analysis that tested the Group (CHs vs. HSs) × Indulgence × Affect in predicting smoking. This interaction test asks whether the difference between CHs and HSs in the relationship of affect and smoking is moderated by being in an indulgent or nonindulgent setting. (In other words, it tests whether the interaction of group and affect, predicting smoking, differs between indulgent and nonindulgent settings. Note that, as expected, indulgent settings are associated with more positive affect, as evident in Figure 5, but this main effect does not bias the evaluation of the interaction of interest.)

The interaction was significant (OR = 1.79, 1.28–2.50, p < .001) and is illustrated in Figure 5, which graphs the mean negative affect score for occasions varying by setting (indulgent vs. not) and smoking, for each group. Analyses within groups were informative. For CHs, there was a strong and significant Indulgence × Affect interaction: In nonindulgent situations, smoking was associated with more negative affect (OR = 1.47, 1.10–1.97, p < .001). Among HSs, there was a modest but significant interaction, but in the opposite direction (OR = 0.84, 0.72–0.98, p < .05): In nonindulgent situations (but not in indulgent situations), smoking was associated with more positive affect. In both cases, affect was related to smoking only outside indulgent settings.

The preceding analyses examined groupwise trends in associations between smoking and situational antecedents. However, these analyses do not completely capture stimulus associations with smoking because they miss idiographic or idiosyncratic associations that vary subject to subject. To capture these, we performed logistic regressions separately for each subject, predicting smoking from situational antecedents. We used the AUC-ROC to quantify the strength of the multivariate association for each within-subject logistic regression and then compared the average strength of association for CHs and HSs by using simple t tests and the nonparametric Wilcoxon's test. Figure 6 shows the resulting average AUC-ROC values for the domains of activity, consumption, affect, social setting, location, and time. CHs' smoking was under significantly greater stimulus control for each category of situational antecedents. (This was true irrespective of whether tested parametrically or nonparametrically and regardless of whether the analysis was weighted.) The association of CHs' smoking with activities was especially evident. Among CHs, knowing current activity would enable one to predict smoking almost 83% of the time (based on an AUC-ROC of 0.83); the parallel figure for HSs was about 65% (recall that 50% correct identification is expected at random). CH-HS differences accounted for 68% of the variance in the AUC-ROC for activities. In view of the fact that there were no systematic associations between affect and smoking in group-aggregate analyses of either group (or the interaction), it is particularly striking that current affect allows one to predict smoking 75% of the time for CHs and 64% of the time for HSs. This illustrates how the AUC-ROC analysis picks up idiographic associations, even when no aggregate group trend is present. Although time of day was related to smoking, these analyses were essentially unchanged when we limited the data to observations collected after 5 p.m.

This is the first study to contrast the smoking patterns of dependent and nondependent smokers. Although we have previously documented HSs' smoking patterns in two studies (Shiffman et al., 2002; Shiffman, Paty, et al., 2004), this is the first study to examine smoking patterns in nondependent smokers. Smoking patterns of CHs selected to demonstrate resistance to nicotine dependence suggested that they were indulgent smokers (M. A. H. Russell et al., 1974) who smoked in relaxing, undemanding situations, and when eating and drinking, but not when working. However, contrary to expectations, they were neither positive affect smokers (affect was not associated with smoking) nor social smokers (they smoked alone often—as often as did the HSs). CHs' smoking patterns distinguished them from HSs, but the nomothetic group differences understated the differences in stimulus control over smoking. When we considered idiographic measures of stimulus control, CHs' smoking proved to be under considerable stimulus control, particularly in association with concurrent activities. In every situational dimension we examined, CHs' smoking was under greater stimulus control than was that of HSs.

One issue in contrasting CHs' and HSs' smoking patterns is that CHs tend to concentrate their smoking later in the day. However, controlling for this by analyzing evening smoking made relatively little difference in the comparisons. This suggests that time of day, per se, does not matter as much in CHs' smoking as do the situations associated with relaxation, eating, drinking, and leisure time, which happen to concentrate in the evening hours. Controlling for time of day did, however, eliminate the initial finding that CHs (more than HSs) tended to smoke when experiencing low arousal; this was apparently due to CHs' tendency to smoke in the evening.

Our data show that CHs are not social smokers (as defined by Moran et al., 2004). If anything, CHs tended to smoke more when alone than with others. CHs smoked almost half their cigarettes when alone—the same as did the HSs. Moreover, even when others were present while CHs were smoking, more often than not, they were not smoking. Less than a quarter of all CHs' cigarettes were smoked in the presence of others who were smoking, and CHs' and HSs' smoking were equally likely to smoke when others were smoking (with no difference in the influence of others' smoking, after time of day was accounted for). Clearly, then, even though CHs sometimes smoked with others, their smoking is not just maintained by a desire to fit in or join other smokers, which suggests it is intrinsically motivated.

The picture of the typical chipper is what M. A. H. Russell et al. (1974) labeled an indulgent smoker—one who smokes when relaxed or socializing, and over meals, coffee, and drinks, but not when working. In fact, three quarters of CHs' cigarettes were smoked under such circumstances. CHs differed from HSs in this respect, though the difference was more quantitative than qualitative: HSs showed similar associations, but they were much weaker and often nonsignificant (Shiffman, Paty, et al., 2004). It is striking that CHs' pattern of smoking appears to resemble patterns associated with early in smokers' careers, before they have progressed to regular daily smoking and to dependence (McKennel & Thomas, 1967; M. A. H. Russell, 1971). This is consistent with our hypothesis that CHs represent smokers who simply have not progressed developmentally into adult forms of dependent smoking (Shiffman, 1991; Shiffman, Paty, et al., 1994).

Despite their pattern of indulgent smoking, CHs' smoking was not associated with positive affect; in general, CHs' affective state did not seem to matter, paralleling our findings for HSs (Shiffman et al., 2002; Shiffman, Paty, et al., 2004). However, when CHs were smoking outside of indulgent situations, smoking was associated with negative affect. This could represent an attempt by CHs to cope with distress in these settings. Whether smoking in fact has any impact on affect cannot be discerned from these data, because we looked at the antecedents of smoking and not the consequences. In contrast to CHs, HSs were not more likely to smoke under negative affect in nonindulgent settings; their smoking was actually slightly more likely to occur under positive affect. It is not clear how to explain these differences. In any case, our analysis suggests and illustrates an analytic strategy of going beyond analysis of overall relationships to partition situations into different categories that may display different smoking dynamics, looking for complex interactions among settings, activities, and affect. This approach has some parallels in studies of relapse episodes, where some episodes are associated with negative affect and others are associated with positive affect, smoking cues, and drinking (Shiffman, 1986), and certain associations with prior negative affect are seen only for a subset of relapse episodes (Shiffman & Waters, 2004).

Our largely negative findings on ad lib smoking and affect (in this and other studies; Shiffman et al., 2002) contrast strikingly with findings on cues that promote smoking under conditions of abstinence, that is, in relapse episodes. In that setting, negative affect appears to play a very powerful role in triggering temptation to smoke and actual smoking (O'Connell & Martin, 1987; Shiffman, 1982; Shiffman, Paty, Kassel, & Hickcox, 1996). During ad lib smoking, smokers may smoke on a regular schedule unconsciously designed to avoid withdrawal, thus blunting the association of smoking and negative affect. Moreover, proprioceptive cues associated with low nicotine levels (Kozlowski & Herman, 1984) or subtle increases in negative affect (Baker, Piper, McCarthy, Majeskie, & Fiore, 2004) might by quite subtle and operate below the level of consciousness, making them unavailable to self-report. During ad lib smoking, these subtle cues may quickly trigger smoking without rising to consciousness. In abstinence, however, smokers are motivated to hold off smoking, and negative affect and craving can build to levels that are not only conscious but also intense. In addition, in abstinence, withdrawal-related negative affect, or withdrawal-related sensitization to negative affect from other sources, may also intensify affect and make it more important and more prominent as a driver of smoking. Finally, negative affect may also become particularly important in relapse because it disrupts smokers' attempts to cope with craving (Muraven & Baumeister, 2000); this dynamic would not influence ad lib smoking.

We also note that the assessment and analysis of affect is controversial (Larsen & Diener, 1992), with some theories viewing negative and positive affect as opposite poles of a single dimension (J. Russell, 1980), as our assessment implies, and others viewing positive and negative affect as independent dimensions (Watson, Clark, & Tellegen, 1988). The difference between the two positions turns on a difference in factor rotation, with a positive affect factor defined by a contrast of positive affect and high arousal versus negative affect and low arousal (Barrett & Russell, 1998; Yik, Russell, & Barrett, 1999). Accordingly, the fact that we failed to find affect by arousal interactions suggests that the findings are not an artifact of the scoring affect according to the circumplex model. It is also notable that we have repeatedly found no relationship between smoking and affect, even when examining individual affect items, suggesting that the finding is not an artifact of how the items are aggregated into factors.

Whereas some laboratory studies have examined how affect may influence how intensely people smoke after they have lit up (e.g., Payne, Schare, Levis, & Colletti, 1991), our study examines only the variables associated with initially lighting up the cigarette. It has been pointed out that the stimuli associated with lighting up itself, and with the effects of the initial nicotine intake from the first puffs (called “intraadministration cues”; McDonald & Siegel, 2004), might come to acquire potent value as conditioned stimuli and may promote continued smoking after smoking has been initiated. Because our method focuses on stimuli that precede, and perhaps trigger, the act of smoking in the first place, these influences (which, in any case, cannot explain why an act of smoking is initiated) were not within the scope of our methods. Likewise, because our focus was on the most immediate antecedents or triggers of smoking, less acute phenomena (e.g., if having a bad day increases smoking over the whole day, even when feeling better) would not be seen in these analyses because the negative affect would pervade both smoking and nonsmoking occasions. Thus, our findings address the stimuli that cue smoking but not other kinds of stimulus influences.

What do these data tell us about CHs' motivation for smoking? We can rule out a number of motives for CHs' smoking. Because CHs do not experience withdrawal (Shiffman, Kassel, et al., 1994) and do not maintain steady-state nicotine levels (Shiffman, Paty, et al., 1994), withdrawal relief cannot be an important motive. The fact that CHs smoke when alone, and are no more likely than HSs to smoke when others are smoking, rules out purely social motives. That CHs do not seek out smoking when working, but instead avoid it, suggests that they are not motivated by performance enhancement. The fact that CHs are most likely to smoke in indulgent situations that are already positively reinforcing suggests that smoking might enhance the reinforcing value of already-reinforcing stimuli. This is consistent with the recent suggestions by Donny et al. (2003), who, in an experiment with rodents taught to self-administer nicotine, found that getting nicotine appeared to make accompanying stimuli reinforcing. In their work, stimuli previously paired with nicotine facilitated self-administration and were sufficient to maintain lever pressing on their own. In other words, nicotine, in addition to being reinforcing in itself, might lend reinforcement value to other stimuli. This suggests that CHs might smoke in circumstances that are already reinforcing to enhance the reinforcement value of those “natural” rewards. This hypothesis should be tested more systematically, both in CHs and in HSs.

This reinforcement-enhancement hypothesis might also explain the very striking relationship between smoking and drinking among CHs (but see Rose et al., 2004). Alcohol consumption was the single best predictor of smoking among CHs. Nevertheless, alcohol can explain only a small fraction of CHs' smoking; only one out of six cigarettes was smoked within 5 min of drinking. Just as CHs are not social smokers, they also are not just drinking smokers.

Our analysis of urges to smoke shows that CHs do express strong desire to smoke at the times they elect to smoke. Indeed, their urge levels at those times are essentially identical to those of HSs. This quantification could be misleading if CHs and HSs were using different scales to assess their craving relative to their range of craving experience. Having never experienced the intense craving that characterizes addictive smokers, CHs might rate cravings as intense that HSs consider to be mild (relatively speaking; see Sayette et al., 2000). Alternatively, CHs may infer craving from the fact that they are smoking (“I am about to smoke, so I must be having an urge”). Taking the craving ratings at face value, the data suggest that CHs and HSs differ dramatically in the frequency and pervasiveness of craving. Whereas the urge intensity associated with smoking occasions was similar for CHs and HSs ( Figure 3, right-hand panel), HSs experienced this heightened craving state every 30 min (crudely, 32 occasions in 16 waking hours), whereas CHs were in this state only about once every 240 min (crudely, 4 occasions in 16 waking hours). Thus, CHs spend much less time in a craving state. Now consider the remaining part of the day, when subjects were not smoking. Whereas HSs were in craving states 53% of the remaining time, CHs were in craving states only 15% of the remaining time. In other words, HSs' craving is pervasive despite a high volume of smoking; CHs' craving is tightly confined despite smoking only rarely.

Whereas craving has typically been linked to dependence, the mere presence of an urge need not implicate dependence processes any more than a social drinker's desire (urge) to have a glass of wine with dinner. Urges become pathological when they become pervasive and become detached from stimulus control. Thus, the difference we see between CHs and HSs is in the frequency and intensity of their urges when they are not smoking. (In other analyses, we found that other measures of dependence correlated best with urge levels when not smoking; Shiffman, Waters, et al., 2004.)

One might say that CHs' cravings demonstrate situational specificity, appearing only in limited situations. The situational specificity of CHs' smoking was similarly demonstrated by our analysis of smoking in each group. The analysis demonstrated that CHs' smoking was under considerable stimulus control, much more so than HSs' smoking. Concurrent activities exercised the greatest influence over CHs' smoking, so much so that knowing what activity a chipper was engaged in would allow one to correctly predict smoking 85% of the time (vs. 67% for HSs). In every stimulus domain—affect, location, activity, eating and drinking, and social setting—CHs' smoking showed significantly more linkage to situational stimuli; it was under greater stimulus control. The fading of stimulus control may well be a hallmark of the development of dependence. A key transition in tobacco use (and other drug use) may occur when users transition from using drugs for particular recreational, or even medicinal, effects in particular contexts to needing and using the drug all the time, regardless of context (M. A. H. Russell et al., 1974). Consistent, regular, and stereotyped patterns of use are hallmarks of dependence (Edwards, 1986; Shadel, Shiffman, Niaura, Nichter, & Abrams, 2000; Shiffman, Waters, & Hickcox, 2004). Thus, the loosening of stimulus control over smoking may be an important process in the development of dependence. We hypothesize that this loosening occurs over time as smokers proceed on a developmental trajectory toward dependence.

This discussion implicitly contrasts two patterns of smoking: one in which smoking is stimulus bound and largely concentrated in indulgent situations and one that is pervasive and relatively free of external stimulus control. We speculate that the situationally limited, indulgent pattern characterizes an early stage in the development of smoking. Smoking is initially triggered by social inducements and, for a time, is socially motivated and tied to social situations (M. A. H. Russell et al., 1974; Shiffman, 1991). As young smokers (like the rodents in Donny's studies) learn that nicotine can enhance the value of other reinforcers, smoking concentrates in indulgent situations where the smoker has access to other reinforcement. CHs, having failed to progress beyond this stage, represent the “fossilized” form of this pattern. Most smokers, however, progress to more frequent and less stimulus-bound smoking, as dependence begins to emerge and smoking comes to be dominated by the motives of craving and withdrawal relief (or avoidance). Research is needed on smoking patterns that emerge as young smokers traverse developmental trajectories of smoking toward dependence (Shadel et al., 2000).

Our interpretation rests on the observation that CHs' smoking showed consistently stronger associations with a variety of situational variables. One might argue that HSs' heavier smoking makes it inherently more difficult to demonstrate associations with situational stimuli; if one is to smoke 30 times a day, one has little discretion about where or when to smoke. However, it is notable that HSs actually showed situational associations as strong as or stronger than did CHs for some variables, such as being in a vehicle or waiting (see Table 2). This demonstrates that even heavy HSs still have “room” to allocate their smoking differentially across contexts, as suggested by the boundary model of nicotine regulation (Kozlowski & Herman, 1984). Conceptually, the fact that heavy, continuous smoking tends to obliterate stimulus associations is in fact consistent with the key point: As a smoker is driven to smoke frequently, stimulus control over the behavior is lost. Conversely, stimulus control must be relaxed to allow smoking to reach the high levels seen in dependent smokers.

Our findings on stimulus control may have implications for prevention and treatment. Our observations on CHs' smoking suggest that stimulus control may be able to act as a strong brake on smoking. Interventions that promote stimulus control of smoking (e.g., restrictions on where one can smoke) could conceivably impede progression toward heavy and dependent smoking. Indeed, it is striking that occasional smoking is more common in states that have strong smoking regulations (Tauras, 2004). Stimulus control interventions might also help promote successful cessation. Cinciripini (Cinciripini, Wetter, & McClure, 1997) has demonstrated that imposing artificial stimulus control on smoking (during a structured reduction phase) can promote cessation, and the data on the stimulus specificity of relapse episodes suggest that exercising stimulus control by avoiding certain situational stimuli could prove effective in preventing relapse. More research is needed on stimulus control of smoking.

Our statistical approach to assessing stimulus control of smoking seemed to provide useful insight into smoking patterns. The analysis of group differences in associations with particular stimuli requires that many of the group members share the same pattern of associations, thus shifting the group means. However, this does not allow for idiographic differences in smoking patterns, which would be expected to affect stimulus control. In our ROC analysis, rather than assuming that smokers in each group would share particular stimulus associations (e.g., that all CHs would smoke more at home), we allowed for idiographic stimulus associations with smoking, such that some individuals might smoke more at home, others when at others' homes, and others at work. The difference in approaches was illustrated by our analysis of affect. When assessed nomothetically, by traditional analysis that required that subjects share similar associations between smoking and particular affects, there were no systematic associations between smoking and affect in either group and no differences between groups. However, the idiographic approach, which allowed associations to vary by individuals, indicated that smoking was under partial control of affective stimuli.

The finding that affect exercises some control over HSs' smoking, when assessed idiographically, may help explain why HSs report that their smoking is prompted by mood changes, even though nomothetic analyses have repeatedly shown no differences in mood between smoking and nonsmoking occasions (Shiffman, 1993; Shiffman, et al., 2002; Shiffman, Paty, et al., 2004); smokers' self-reports may pick up on the idiographic associations between smoking and mood. However, the magnitude of the associations between mood and smoking seen even in idiographic analyses does not approach that which smokers report on global questionnaires (Shiffman, 1993), suggesting that there is still a discrepancy between global retrospective reports and the detailed self-monitoring in this study. The fact that subjectively experienced affect plays such a prominent role in relapse (Shiffman et al., 1996; Shiffman & Waters, 2004) may leave smokers with a broad impression that their smoking is linked to negative affect.

In any case, the study suggests that it is important to allow for idiographic associations in analyzing smoking patterns, particularly because stimulus associations with smoking might develop through accidental conditioning to certain stimuli and thus should be expected to vary across individuals, perhaps even randomly.

Although we have made considerable progress in understanding dependent smoking, relatively little attention has been paid to understanding nondependent or situation-bound smoking. Although heavy and dependent smoking is still the predominant smoking pattern in the United States and Western Europe, it may be important, for several reasons, to understand the patterns, motives, and reinforcers that drive and maintain low-level smoking. First, although heavier smoking leads to increased health risks (U.S. Department of Health and Human Services, 1998), even low-level smoking has adverse health consequences; Luoto (Luoto, Uutela, & Puska, 2000) reported that even nondaily smoking increases the risk of heart disease by 50%. Second, contrasting these patterns of smoking with more typical heavy or dependent smoking can help improve our understanding of how smoking develops and is maintained. Finally, low-level and occasional smoking is becoming an increasingly common pattern, with 24% of U.S. smokers engaging in less-than-daily smoking, an increase of almost 50% between 1996 and 2001 (Centers for Disease Control, 2003). Our sample of CHs was selected to represent an extreme of resistance to dependence (i.e., we recruited smokers who maintained low levels of smoking despite smoking on most days over many years) and thus was not representative of this growing group, many of whom smoke at lower levels and with less stability than seen in our CHs sample (Gilpin, Cavin, & Pierce, 1997). Studies of smoking patterns among the full representative range of intermittent or nondependent smokers are needed.

Our study's conclusions are bounded by certain limitations. The study was done at a time when restrictions on smoking were relatively modest (see Shiffman, Paty, et al., 2004). The current climate of smoking regulations may have imposed some stimulus control on smokers while disrupting other associations (see Shiffman et al., 2002). Our samples of CHs and HSs were small and arguably not representative. Very stringent criteria were applied for inclusion of CHs, and HSs were systematically selected to match the CHs sample. The validity of the results also depends on the validity of the EMA data on smoking situations. To the extent that these were biased by reactivity or by some kinds of systematic noncompliance (e.g., not recording cigarettes when very upset), the data could be misleading. However, analyses suggested that compliance was very good and that the data were valid. The use of EMA methods, relying on real-time entry of real-world data regarding smoking situations, and the contrast to randomly sampled nonsmoking situations, was a significant strength of our methodology. The large number of observations per person may help compensate for the small sample of persons, and the observation of each person over 2 weeks of ad lib smoking helped ensure that each person's smoking patterns were adequately sampled. Finally, the use of a palmtop computer to collect the data helped ensure that data were recorded in a timely manner (Stone et al., 2002) and enabled analysis of temporal patterns.

This study documents the smoking patterns of nondependent smokers and contrasts them to those of HSs. The study demonstrates that CHs are not social smokers but that their smoking is concentrated in situations associated with relaxation and consumption of food, coffee, and alcohol. An idiographic analysis demonstrated that CHs' smoking was under stronger stimulus control than was HSs' smoking, with specific patterns of stimulus associations being individual and idiosyncratic. Understanding how smoking is brought under stimulus control, and how it breaks out of such control as dependence develops, is an urgent agenda for the study of smoking and the development of nicotine dependence, with implications for prevention and cessation interventions.

1  The term stimulus control is commonly used to describe stimulus-behavior contingencies. As our analyses are observational and do not involve manipulation of antecedent stimuli to observe their effect on smoking, the causal implications of “control” may not apply. However, we use the term to link to the existing literature on stimulus control, including stimulus control of smoking.

2  The analysis follows a case-control design, in which the antecedents of cases (smoking episodes) and controls (nonsmoking episodes) are described. From these data, it is not possible to say how likely a subject is to smoke, for example, when working, but it is possible to say that the likelihood of smoking is increased when working and to quantify, on the basis of the odds ratio, the degree to which the odds of smoking are increased (Rudas, 1998).

3  Conceptually, then, the AUC-ROC expresses the overall predictability of smoking, as does the R2 statistic in a linear regression. An alternative statistic that has been offered as an analogue for R2 in logistic regression is the reduction in the log-likelihood statistic due to the model fit (essentially, the reduction in uncertainty; Menard, 1995). The two statistics capture similar information and were very highly correlated (average = 0.95) in our data. We also tested our models with the reduction in log-likelihood and observed the same results. We prefer the AUC-ROC statistic for presentation because of its interpretation as the probability of correct identification.

4  HSs contributed more smoking assessments because we collected five smoking assessments per day, whereas most CHs did not smoke five times daily. Regular smokers had fewer random assessments than did CHs because they smoke more and spend less time not smoking, thus providing fewer opportunities for random assessments.

5  This explanation is made less plausible by the fact that exploratory analyses reveal that CHs' craving (as observed in randomly scheduled assessments) starts to rise about a half hour or so before they smoke.

6  It is not necessary to posit that the original pattern of smoking in indulgent situations disappears; rather, it is covered over by more pervasive smoking patterns. This explains why similar situational associations are seen in HSs in a much weaker, more diluted form.

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.

Barrett, L. F., & Russell, J. A. (1998). Independence and bipolarity in the structure of current affect. Journal of Personality and Social Psychology, 74, 967–984.

Brauer, L. H., Hatsukami, D., Hanson, K., & Shiffman, S. (1996). Smoking topography in tobacco chippers and dependent smokers. Addictive Behaviors, 21, 233–238.

Bryk, A. S., & Raudenbush, S. W. (1992). Hierarchical linear models: Applications and data analysis methods. Newbury Park, CA: Sage.

Centers for Disease Control. (2003). Prevalence of current cigarette smoking among adults and changes in prevalence of current and some day smoking: United States, 1996–2001. Morbidity and Mortality Weekly Report, 52, 303–307.

Chassin, L., Presson, C. C., & Sherman, S. J. (1984). Cognitive and social influence factors in adolescent smoking cessation. Addictive Behaviors, 9, 383–390.

Cinciripini, P. M., Wetter, D. W., & McClure, J. B. (1997). Scheduled reduced smoking: Effects on smoking abstinence and potential mechanisms of action. Addictive Behaviors, 22, 759–767.

Csikszentmihalyi, M., & Larson, R. (1987). Validity and reliability of the experience-sampling method. Journal of Nervous and Mental Disease, 175, 509–513.

Davies, G. M., Willner, P., & Morgan, S. F. (2000). Smoking-related cues elicit craving in tobacco “chippers”: A replication and validation of the two-factor structure of the Questionnaire of Smoking Urges. Psychopharmacology, 152, 334–342.

Donny, E. C., Chaudhri, N., Caggiula, A. R., Evans-Martin, F. F., Booth, S., Gharib, M. A., et al. (2003). Operant responding for a visual reinforcer in rats is enhanced by noncontingent nicotine: Implications for nicotine self-administration and reinforcement. Psychopharmacology, 169, 68–76.

Edwards, G. (1986). The alcohol dependence syndrome: A concept as stimulus to enquiry. British Journal of Addiction, 81, 171–183.

Gilpin, E., Cavin, S., & Pierce, J. (1997). Adult smokers who do smoke daily. Addiction, 92, 473–480.

Gilpin, E. A., Emery, S. L., Farkas, A. J., Distefan, J. M., White, M. M., & Pierce, J. P. (2001). The California Tobacco Control Program: A decade of progress. Results from the California Tobacco Surveys, 1990–1998. La Jolla, CA: University of California, San Diego.

Hajek, P., West, R., & Wilson, J. (1995). Regular smokers, lifetime very light smokers, and reduced smokers: Comparison of psychosocial and smoking characteristics in women. Health Psychology, 14, 195–201.

Hammersley, R. (1994). A digest of memory phenomena for addiction research. Addiction, 89, 283–293.

Hanley, J. A., & McNeil, B. J. (1982). The meaning and use of the area under a receiving operating characteristic (ROC) curve. Radiology, 143, 29–36.

Istvan, J., & Matarazzo, J. D. (1984). Tobacco, alcohol and caffeine use: A review of their interrelationships. Psychological Bulletin, 95, 301–326.

Jarvik, M., Killen, J. D., Varady, A., & Fortman, S. P. (1993). The favorite cigarette of the day. Journal of Behavioral Medicine, 16, 413–422.

Kozlowski, L. T., & Herman, C. P. (1984). The interaction of psychosocial and biological determinants of tobacco use: More on the boundary model. Journal of Applied Social Psychology, 14, 244–256.

Larsen, R. J., & Diener, E. (1992). Promises and problems with the circumplex model of emotion. In M.Clark (Ed.), Review of personality and social psychology (pp. 25–59). Newbury Park, CA: Sage.

Luoto, R., Uutela, A., & Puska, P. (2000). Occasional smoking increases total and cardiovascular mortality among men. Nicotine and Tobacco Research, 2, 133–139.

McDonald, R. V., & Siegel, S. (2004). Intra-administration associations and withdrawal symptoms: Morphine-elicited morphine withdrawal. Experimental and Clinical Psychopharmacology, 12, 3–11.

McKee, S. A., Hinson, R., Roundsaville, D., & Petrelli, P. (2004). Survey of subjective effects of smoking while drinking among college students. Nicotine and Tobacco Research, 6, 111–117.

McKennel, A. C., & Thomas, R. K. (1967). Adults' and adolescents' smoking habits and attitudes. London: British Ministry of Health.

Menard, S. W. (1995). Applied logistic regression analysis (Series No. 07–106). Sage University Paper Series on Quantitative Applications in the Social Sciences. Thousand Oaks, CA: Sage.

Moran, S., Wechsler, H., & Rigotti, N. A. (2004). Social smoking among U.S. college students. Pediatrics, 114, 1028–1034.

Muraven, M., & Baumeister, R. F. (2000). Self-regulation and depletion of limited resources: Does self-control resemble a muscle?Psychological Bulletin, 126, 247–259.

O'Connell, K. A., & Martin, E. J. (1987). Highly tempting situations associated with abstinence, temporary lapse, and relapse among participants in smoking cessation programs. Journal of Consulting and Clinical Psychology, 55, 367–371.

Paty, J. A., Kassel, J. D., & Shiffman, S. (1992). Assessing stimulus control of smoking: The importance of base rates. In M.DeVries (Ed.), The experience of psychopathology: Investigating mental disorders in their natural setting (pp. 347–352). Cambridge, England: Cambridge University Press.

Payne, T. J., Schare, M. L., Levis, D. J., & Colletti, G. (1991). Exposure to smoking-relevant cues: Effects on desire to smoke and topographical components of smoking behavior. Addictive Behaviors, 16, 467–479.

Rose, J. E., Brauer, L. H., Behm, F. M., Cramblett, M., Calkins, K., & Lawhon, D. (2004). Psychopharmacological interactions between nicotine and ethanol. Nicotine & Tobacco Research, 6, 133–144.

Rudas, T. (1998). Odds ratios in the analysis of contingency tables. Thousand Oaks, CA: Sage.

Russell, J. (1980). A circumplex model of affect. Journal of Personality and Social Psychology, 37, 345–356.

Russell, M. A. H. (1971). Cigarette smoking: The natural history of a dependence disorder. British Journal of Medical Psychology, 44, 1–16.

Russell, M. A. H., Peto, J., & Patel, U. A. (1974). The classification of smoking by factorial structure of motives. Journal of the Royal Statistical Society, 137, 313–346.

Sayette, M. A., Shiffman, S., Tiffany, S. T., Niaura, R., Martin, C. S., & Shadel, W. G. (2000). The measurement of drug craving. Addiction, 95, (Suppl. 2)S189–S210.

Sayette, M. A., Wertz, J. M., Martin, C. S., Cohn, J. F., Perrott, M. A., & Hobel, J. (2003). The effect of smoking opportunity on cue-elicited urge: A facial code analysis. Experimental and Clinical Psychopharmacology, 11, 218–227.

Sayette, M. A., Wilson, G. T., & Elias, M. J. (1992). Alcohol and aggression: A social information processing analysis. Journal of Studies on Alcohol, 54, 399–407.

Shadel, W. G., Shiffman, S., Niaura, R., Nichter, M., & Abrams, D. B. (2000). Current models of nicotine dependence: What is known and what is needed to advance understanding of tobacco etiology among youth. Drug and Alcohol Dependence, 59-S1, S9–S22.

Shiffman, S. (1982). Relapse following smoking cessation: A situational analysis. Journal of Consulting and Clinical Psychology, 50, 71–86.

Shiffman, S. (1986). A cluster-analytic classification of smoking relapse episodes. Addictive Behaviors, 11, 295–307.

Shiffman, S. (1989). Tobacco “chippers”: Individual differences in tobacco dependence. Psychopharmacology, 97, 539–547.

Shiffman, S. (1991). Refining models of dependence: Variations across persons and situations. British Journal of Addiction, 86, 611–615.

Shiffman, S. (1993). Assessing smoking patterns and motives. Journal of Consulting and Clinical Psychology, 61, 732–742.

Shiffman, S. (2004). Designing protocols for ecological momentary assessment. In A. A.Stone, S.Shiffman, & A.Atienza (Eds.), The science of real-time data capturing: Self-reports in health research. New York: Oxford University Press.

Shiffman, S., & Balabanis, M. (1995). Associations between alcohol and tobacco. In J. B.Fertig & J. P.Allen (Eds.), Alcohol and tobacco: From basic science to clinical practice (pp. 17–36). Washington, DC: U. S. Government Printing Office.

Shiffman, S., Fischer, L. A., Zettler-Segal, M., & Benowitz, N. E. (1990). Nicotine exposure in non-dependent smokers. Archives of General Psychiatry, 47, 333–336.

Shiffman, S., Gwaltney, C. J., Balabanis, M. H., Liu, K. S., Paty, J. A., Kassel, J. D., et al. (2002). Immediate antecedents of cigarette smoking: An analysis from ecological momentary assessment. Journal of Abnormal Psychology, 111, 531–545.

Shiffman, S., Hufford, M., Hickcox, M., Paty, J. A., Gnys, M., & Kassel, J. D. (1997). Remember that? A comparison of real-time versus retrospective recall of smoking lapses. Journal of Consulting and Clinical Psychology, 65, 292–300.

Shiffman, S., Kassel, J. D., Paty, J. A., Gnys, M., & Zettler-Segal, M. (1994). Smoking typology profiles of chippers and regular smokers. Journal of Substance Abuse, 6, 21–35.

Shiffman, S., & Paton, S. M. (1999). Individual differences in smoking: Gender and nicotine addiction. Nicotine and Tobacco Research, 1, S153–S157.

Shiffman, S., Paty, J. A., Gnys, M., Kassel, J. D., & Elash, C. (1995). Nicotine withdrawal in chippers and regular smokers: Subjective and cognitive effects. Health Psychology, 14, 301–309.

Shiffman, S., Paty, J. A., Gwaltney, C. J., & Dang, Q. (2004). Immediate antecedents of cigarette smoking: An analysis of unrestricted smoking patterns. Journal of Abnormal Psychology, 113, 166–171.

Shiffman, S., Paty, J. A., Kassel, J. D., Gnys, M., & Zettler-Segal, M. (1994). Smoking behavior and smoking history of tobacco chippers. Experimental and Clinical Psychopharmacology, 2, 126–142.

Shiffman, S., Paty, J. A., Kassel, J. D., & Hickcox, M. (1996). First lapses to smoking: Within-subjects analysis of real-time reports. Journal of Consulting and Clinical Psychology, 62, 366–379.

Shiffman, S., & Prange, M. (1988). Self-monitored and self-reported smoking patterns: A validity study. Addictive Behaviors, 13, 201–204.

Shiffman, S., & Waters, A. J. (2004). Negative affect and smoking lapses: A prospective analysis. Journal of Consulting and Clinical Psychology, 72(2), 192–201.

Shiffman, S., Waters, A. J., & Hickcox, M. (2004). The Nicotine Dependence Syndrome Scale: A multi-dimensional measure of nicotine dependence. Nicotine and Tobacco Research, 6, 327–348.

Shiffman, S., Zettler-Segal, M., Kassel, J. D., Paty, J. A., Benowitz, N. L., & O'Brien, G. (1992). Nicotine elimination and tolerance in nondependent cigarette smokers. Psychopharmacology, 109, 449–456.

Stone, A. A., & Shiffman, S. (1994). Ecological momentary assessment (EMA) in behavioral medicine. Annals of Behavioral Medicine, 16, 199–202.

Stone, A. A., Shiffman, S., Schwartz, J. E., Broderick, J. E., & Hufford, M. R. (2002). Patient non-compliance with paper diaries. British Medical Journal, 324, 1193–1194.

Tate, J. C., Schmitz, J. M., & Stanton, A. L. (1991). A critical review of the Reasons for Smoking Scale. Journal of Substance Abuse, 3, 441–445.

Tauras, J. A. (2004). Public policy and some-day smoking among adults. Journal of Applied Economics, 7, 137–162.

U.S. Department of Health and Human Services. (1988). The health consequences of smoking: Nicotine addiction. A report of the Surgeon General. Washington, DC: Author.

Vogt, T. M., Selvin, S., Widdowson, G., & Hulley, S. B. (1977). Expired air carbon monoxide and serum thiocyanate as objective measures of cigarette exposure. American Journal of Public Health, 67, 545–549.

Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and validation of brief measures of positive and negative affect: The PANAS scales. Journal of Personality and Social Psychology, 54, 1063–1070.

Yik, M. S. M., Russell, J. A., & Barrett, L. F. (1999). Structure of self-reported current affect: Integration and beyond. Journal of Personality and Social Psychology, 77, 600–619.

Zeger, S. L., Liang, K., & Albert, P. S. (1988). Models for longitudinal data: A generalized estimating equation approach. Biometrics, 44, 1049–1060.

Zhu, S., Sun, J., Hawkins, S., Pierce, J., & Cunningham, J. (2003). A population study of low-rate smokers: Quitting history and instability over time. Health Psychology, 22, 245–252.

Submitted: September 3, 2004 Revised: May 23, 2005 Accepted: May 24, 2005