Associations between Morning Affect and Later-Day Smoking Urges and Behavior

Abstract

Objective:

Affective experiences are associated with smoking urges and behavior. Few studies have examined the temporal nature of these associations within a day, such as whether positive and negative affect in the morning are associated with smoking urges and behavior later in the day.

Method:

Participants (N=63; M_Age=50 years, 48% Female; 60% White) were randomized into one of three smoking cessation interventions and answered up to 5 daily ecological momentary assessments for 28 days during a quit attempt (M=21.0 days, SD=7.1). Before analysis, scores for morning positive and negative affect and later-day smoking urges and behavior were calculated.

Results:

On days when individuals’ morning positive affect was higher than usual, later-day smoking urges tended to be lower than usual. In contrast, on days when individuals’ morning negative affect was higher than usual, later-day smoking urges tended to be higher than usual, and smoking was more likely. Further, individuals who had higher characteristic morning positive affect tended to have less intense later-day smoking urges, whereas those who tended to have higher characteristic morning negative affect tended to have more intense later-day smoking urges.

Conclusions:

Morning positive and negative affect were associated with later-day smoking urges and morning negative affect was related to later-day smoking behavior. Future research should examine whether interventions that boost positive affect on mornings when it is lower than usual, and attenuate negative affect on mornings when it is higher than usual, may reduce the intensity of smoking urges and the likelihood of smoking later in the day.

Despite decreasing trends in the rate of cigarette smoking, approximately 30.8 million people in the United States smoke cigarettes (Cornelius et al., 2022). Successfully quitting smoking increases life expectancy by a decade or more and decreases smoking-related morbidity and mortality (Jha et al., 2013; Babb et al., 2017). However, once individuals become dependent on nicotine, quitting becomes difficult partly due to withdrawal symptoms such as smoking urges (used synonymously with smoking cravings), as well as heightened negative affect and reduced positive affect (Moore et al., 2014; Potter et al., 2021; Benowitz, 2010). Although many individuals express interest in quitting smoking and even engage in one or more quit attempts, the rate of successful cessation is only ~8%, suggesting room for improvement (US Department of Health and Human Services, 2020). In order to better understand the experiences of adults who are trying to quit smoking, smoking behavior and symptoms of tobacco withdrawal can be assessed repeatedly in real time via ecological momentary assessments (EMAs) delivered through a smartphone app.

Data collected via EMA are ideally suited for capturing the temporal and contextual dynamics of smoking cessation as individuals go about their daily lives. Advantages of EMA include the ability to evaluate both between- and within-person associations, and the extent of between-person differences in within-person associations (Bolger & Laurenceau, 2013). For adults actively trying to quit smoking, it is more common to oscillate between periods of non-smoking and smoking than to completely quit smoking (Chaiton et al., 2016; Shiffman et al., 2007; US Department of Health and Human Services, 2020). Furthermore, during a quit attempt, individuals tend to experience increases in the overall intensity of smoking urges, as well as more extreme fluctuations in smoking urges during the day (Piper et al., 2011). A related study demonstrated that among individuals receiving a smoking cessation intervention, the majority of the variation (81%) in smoking urges was at the momentary assessment level (rather than the person-level), indicating the importance of information that is captured within relatively short time intervals such as minutes or hours (Shiyko & Ram, 2011). Altogether, these studies indicate that smoking urges and behavior vary across relatively fast time-scales for adults who are trying to quit smoking, and demonstrate the critical need to identify and ultimately intervene upon antecedents of smoking behavior at specific time-scales.

Theory and prior research suggest that negative and positive affect are associated with smoking (Williams et al., 2018; Businelle et al., 2010; Businelle et al., 2013). In brief, negative affect is defined as subjective feeling states with a negative valence (e.g., frustrated, sad), while positive affect is defined as subjective feeling states with positive valence (e.g., happy, calm). The Affective Processing Model of Negative Reinforcement suggests that smoking is reinforced through continuous motivation to avoid or alleviate negative affect (Baker et al., 2004). The model further postulates that negative reinforcement processes are generalized to negative affect specific to tobacco withdrawal and negative affect experienced in other daily life contexts. Finally, the model proposes that moments of heightened affective distress may trigger both smoking urges and smoking behavior (Baker et al., 2004).

Consistent with the affective processing model of negative reinforcement, previous research provides evidence of within-person associations between negative affect, and smoking urges and smoking behavior (for a review, see Serre et al., 2015). Relevant to the present study, prior research has indicated that negative affect is an important within-day antecedent of subsequent smoking urges among individuals actively trying to quit smoking. For example, in one study restricted to participants who achieved initial 24-hour abstinence and to the period where abstinence was maintained, higher negative mood (assessed via a bipolar scale ranging from extremely positive to extremely negative), was associated with greater odds of experiencing high smoking urges across time periods spanning three minutes to seven hours (Shiyko et al., 2013). Another study found that for momentary occasions where anxious mood (bipolar scale calm/anxious) was higher than usual, smoking urges approximately four hours later also tended to be higher than usual, after accounting for concurrent smoking urges (Serre et al., 2018).

Several studies have also found that negative affect is associated with subsequent smoking behavior when individuals are trying to quit smoking. For example, Berkman et al. (2011) found that on occasions when participants’ prior mood was more negative than usual for that day (assessed via a bipolar scale), the likelihood of smoking on the next occasion within that day was higher than usual. Similarly, Shiffman and Waters (2004) found that negative affect (assessed via a bipolar scale) tended to increase across several hours preceding participants’ first post-quit smoking episode, particularly for smoking episodes precipitated by a stressor. Three other studies quantified negative affect as the composite of several discrete negative emotions, and found that negative affect was related to higher odds of smoking by the next occasion within a given day (Businelle et al., 2016b; Moore et al., 2014; Potter et al., 2021).

While the affective processing model of negative reinforcement does not make predictions about relations between positive affective experiences and subsequent smoking, some have suggested that positive affect may be a protective or buffering factor tied to lower smoking urges and lower likelihood of smoking. For example, the dynamic model of affect suggests that positive affect is a unique resource that can be mobilized in stressful situations (Zautra et al., 2001). Borland (1991) has also suggested that higher positive affect may facilitate a quicker return to abstinence after a smoking episode (Borland, 1990). A meta-analysis of laboratory-based studies showed that positive affect manipulations were not associated with smoking urges (Heckman et al., 2013). However, it is possible that this finding may not generalize to individuals’ daily life experiences. From the few EMA-based studies of individuals who were actively trying to quit smoking, evidence suggests that positive affective experiences are associated with subsequent smoking within a given day. For example, one study showed higher momentary positive affect was associated with lower odds of smoking by the next occasion within a given day, specifically for days 3 through 11 post-quit (Potter et al., 2021). Other studies found evidence of prospective associations between positive affect and next-day smoking (Ferguson et al., 2006; Vinci et al., 2017). For example, Ferguson et al. (2006) found evidence of a prospective association between positive affect on a given day and lower risk of smoking the next day. Further, when both positive affect and negative affect were entered into the same model, only positive affect explained unique variation in the risk of smoking the next day. The authors suggested that reductions in positive affect may undermine individuals’ resistance to temptation and increase motivation to smoke in order to restore affective balance.

Altogether, these studies provide evidence of within-day prospective associations between negative affect and subsequent smoking urges, and between affect (positive and negative) and subsequent smoking behavior over time-scales ranging from several minutes to several hours. However, research to date has not explicitly addressed whether negative affect in the morning is associated with subsequent smoking urges and smoking behavior later in the day. Additionally, research has not explicitly addressed prospective associations between positive affect and subsequent smoking urges nor whether affect in the morning (positive and negative affect) is associated with subsequent smoking urges and smoking behavior later in the day. Better understanding of whether morning affective experiences are associated with later-day smoking urges and likelihood of later-day smoking has implications for future intervention development. For example, if prospective associations are found between morning affective experiences and later-day smoking urges and smoking likelihood, this would provide impetus for examining effectiveness of an intervention that intervenes when higher than usual negative affect is detected in the morning.

Present Study

The present study is a secondary analysis of data collected as part of a pilot study designed to examine intervention group differences in smoking cessation (see Methods section for further details as well as Hébert et al., 2020 for main outcomes paper). The purpose of the present study is to use EMA data to assess whether individuals’ affective states in the morning operate as protective or risk factors for later-day smoking urges and smoking behavior during a quit attempt. First, we examined whether on mornings when individuals’ positive affect was higher than their usual, their smoking urges later in the day were lower than usual. Similarly, we examined whether on mornings when individuals’ negative affect was higher than their usual, their smoking urges later in the day were also higher than usual. Second, we examined whether on mornings when individuals’ positive affect was higher than usual, their likelihood of smoking later in the day was lower than usual, and whether on mornings when individuals’ negative affect was higher than their usual, their likelihood of smoking later in the day was higher than usual. This information will inform future intervention development and refinement by identifying proximal associations between affect and smoking experiences, and by identifying a timeframe (morning time, shortly after waking) when intervention deployment may be particularly useful.

Method

Transparency and Openness

In the following sections, we report how we determined sample size, data exclusions, manipulations, and measures relevant to the present study. We follow the Journal Article Reporting Standards (JARS; Appelbaum et al., 2018). Data and research materials are not openly available at this time. However, analysis code is available on OSF (https://osf.io/78ngp/). All analyses were performed in R version 4.1.1. and R Studio version 2021.9.0.351 using the brms package (R Core Team, 2020; RStudio Team, 2021; Bürkner, 2018). All figures were created in R using the ggplot2 package (Wickham, 2016). This study’s design and its analyses were not pre-registered. The ClinicalTrials.gov identifier is NCT02930200.

Participants

Eighty-one individuals participated in the parent study (see Hébert et al., 2020 for main outcomes). Of these participants, 18 were excluded (see data analysis section for further details). Participants in the analytic sample (n_Total=63; n_Smart-T2=19, n_QuitGuide=22, n_TTRP=22) were, on average, 50 years of age (SD=11.9, Range= 23-73 years), 48% female, 60% White (21% Black or African American, 6% American Indian/ Alaska Native, 5% Multiracial, 2% Native Hawaiian or Other Pacific Islander, and 6% Other). Three percent identified as Hispanic/Latino. Participants were recruited from provider- or self-referrals to the Tobacco Treatment Research Program (TTRP) at the University of Oklahoma Health Sciences Center. Eligibility included (1) English literacy level greater than the 6th grade, (2) willingness to quit smoking 7 days from the first visit, (3) ≥18 years of age, (4) expired carbon monoxide level >7 ppm suggestive of current smoking (Benowitz et al., 2020), (5) currently smoking ≥5 cigarettes per day, (6) willingness and ability to attend 4 in-person assessment sessions, and (7) no contraindications for over-the-counter nicotine replacement therapy (NRT; e.g., uncontrolled blood pressure, myocardial infarction within the past 2 weeks, or current pregnancy or plans to become pregnant during the study period). Prior to enrollment in the study, participants provided informed consent.

Procedure

All study procedures were approved by the Institutional Review Board at the University of Oklahoma Health Sciences Center. Participants completed a baseline in-person assessment and received training from study personnel regarding use of the study-provided smartphone (Samsung Galaxy On5) to complete daily EMAs. Smartphones were programmed to prompt EMAs only during self-reported typical wake times. During the baseline visit, eligible participants were randomized into either the (1) Smart-T2 smartphone-based smoking cessation app (SmartT2; Businelle et al., 2016a; Hébert et al., 2020), (2) National Cancer Institute QuitGuide smartphone-based smoking cessation app (QuitGuide; https://smokefree.gov/apps-quitguide; Buller et al., 2014), or (3) standard tobacco cessation clinic care which included 6 weekly counseling sessions and 12 weeks of NRT (TTRP; Fiore et al., 2008). All participants were offered an initial 4-week supply of nicotine replacement therapy (NRT; patches, gum, or both), and six additional weeks of NRT if needed. We included intervention type as a covariate, but intervention group differences were not the primary focus of the present study.

After the baseline assessment, participants completed a 7-day pre-quit EMA period, followed by a quit date in-person assessment, and 29 days of post-quit EMAs. Only post-quit EMA data were used for the present study, given our interest in understanding how affective experiences related to smoking urges and behavior during a scheduled quit attempt. On day 29, participants were prompted to respond to a single daily diary assessment. On all other days, participants were prompted five times per day at approximately even intervals during the period between their self-reported wake time and bedtime for each day of the week. The first prompted EMA in the morning (morning assessment) was a daily diary where participants answered questions about their momentary experiences and experiences from the previous day. The four subsequent prompted EMAs were termed random assessments where participants only answered questions about their momentary experiences. In addition to the scheduled EMAs, participants could self-initiate EMAs when they felt an urge to smoke or when they had already smoked. Self-initiated EMAs were only used to calculate whether a participant smoked on a given day.

Measures

Morning Affect

During the morning assessment (i.e., ~30 minutes after waking), participants rated the extent to which they were experiencing several discrete emotions, generally selected in accordance with Russell’s (1980) circumplex model that splits emotion experiences into quadrants reflecting combinations of valence (positive, negative) and arousal (low, high). Responses ranged from Strongly Disagree (=1), Disagree (=2), Neutral (=3), Agree (=4), Strongly Agree (=5). Although fewer discrete positive emotions were assessed compared with discrete negative emotions, the positive affect and negative affect composite scores (described below) each include at least one low arousal emotion and one high arousal emotion. Rationale for this data collection choice was based on prior research that used machine learning, which showed that more negative valence emotions (i.e., irritable, sad, stressed) were among the strongest predictors of imminent smoking lapse compared with positive valence emotions (i.e., calm; Suchting et al., 2019)

Negative Affect

Negative emotions included stress, anxious, frustrated/angry, irritable, worried, restless, depressed, sad, and miserable. The average within-person correlations among the negative emotion items were moderate to strong during the morning assessments in the post-quit period (the averages of the within-person correlations for each bivariate pair ranged from 0.28 to 0.58; M=0.43, SD=0.06). Thus, a morning negative affect score was computed as the average of the nine negative affect items listed above for each morning assessment.

Positive Affect

Positive emotions: happy and calm. The average within-person correlations among the positive emotion items were moderate to strong during the morning assessments in the post-quit period (the average of the within-person correlations was 0.32). Thus, a morning positive affect score was computed as the average response to the two positive emotion items for each morning assessment.

Later-Day Smoking Urges

During each random assessment (prompted after the morning assessment each day), participants reported on the extent to which they currently had an urge to smoke. Response options ranged from Strongly Disagree (=1), Disagree (=2), Neutral (=3), Agree (=4), Strongly Agree (=5). A later-day smoking urge score was calculated for each person on each day by averaging responses across all random assessments for that day after the morning assessment.

Later-Day Smoking Behavior

Within each day, across the morning daily diary, random assessments and participant initiated smoking and urge assessments, participants reported on their momentary smoking behavior for that day (e.g., Today how long ago did you last smoke a cigarette? and How many cigarettes did you just smoke?). Additionally, during the morning daily diary, participants reported on their daily smoking behavior for the prior day. For this variable, scores were lagged by 1 day to reflect current-day smoking. On days when participants did not report (via only the momentary assessments) smoking on or prior to the morning assessment, later-day smoking was coded as yes (=1) if the participant indicated across any of the momentary or day-level assessment types that they had smoked that day, or coded as no (=0) if smoking was not reported during any assessments that day.

Covariates

Demographic Variables

Demographic variables include: Gender (Male=0, Female=1), Race (0=Non-White, 1=White, given that the majority of the sample reported their race as White), Age (years), and Education (completed ≤ high school/GED=0, at least some college=1).

Baseline Smoking Dependence

The Heaviness of Smoking Index (HSI; Heatherton et al., 1989) includes two items pertaining to (1) how long after waking is a first cigarette smoked and (2) how many cigarettes smoked per day. As per standard practice, scores from each item were summed to create a baseline smoking dependence score for each participant.

Intervention Type

Two dummy coded variables were created to indicate whether the participant received the QuitGuide app (0=no, 1=yes) or the TTRP clinic-based care (0=no, 1=yes), with SmartT2 as the reference group in all analyses.

Days Since Quit Date

Days since quit date is a continuous integer variable ranging from 0 (=quit date) to 27 (=27 days post-quit).

Data Analysis

Data Preparation

We performed several initial data preparation steps to examine the present study research questions pertaining to within- and between-person associations among morning affect and later day smoking urges and smoking behavior. First, we defined morning as the period between 5:00 AM and 11:59 AM to capture participants’ self-reported wake-up times and removed any assessment before 5:00 AM (t=198; M=2.6 assessments per person, SD=7.1). Second, we removed 380 assessments (M=5 assessments/person, SD=8.4) across days (t=158 total days; M=2.1 days/person, SD=3.1) where the minimum EMA timestamp was after 12 pm. Third, we removed days when participants smoked before or during the morning assessment (t=271 total days; M=3.6 days/person, SD=6.5). Fourth, we removed days with only a morning observation and no later day data (t=61 total days; M=1 day/person, SD=0.9). Altogether, 18 participants were excluded from the analysis for (1) not providing any post-quit data (n=4), (2) not providing any assessments in the morning period between 5:00 AM and 12:00 PM (n=2), (3) for smoking prior to or at the time of the morning assessment on all study days (n=3), (4) not providing any assessments in the later-day period between 12:00 PM and 11:59 PM (n=1), and (5) not providing at least 5 days with complete data for both morning affect and later day smoking and smoking urges (n=8), in line with prior suggestions for the minimum number of occasions for conducting intensive longitudinal data analysis (Bolger & Laurenceau, 2013).

We operationalized the morning assessment as the earliest assessment provided by the participant between 5:00 AM and 11:59 AM. Additionally, we operationalized the later-day assessment based on pooling information from all other assessments obtained in that same day. Later-day assessments could occur within the morning time span, but were always at least 1 or more hours after the morning assessment. For example, if a participant completed an assessment at 8:00 AM and another assessment at 11:00 AM, the first assessment would be considered the morning assessment and the second assessment would be used as part of the calculation of the later-day score (along with any other later day assessments that the participant responded to). For later-day smoking urges, we calculated the average of all smoking urge scores provided within a given day (except for the morning assessment). For later-day smoking behavior, we determined whether smoking was reported in any momentary-level or day-level assessments. The resulting data included an average of 21 days (SD=7.1) per person during the 29-day post-quit period where no smoking had occurred on or before the morning assessment, and data were available for both morning and later-day. On average, morning assessments occurred at approximately 8:15 AM (SD=1.7). On average, the later-day score for each person on each day was calculated from 3.44 post-morning assessments (SD=0.8), with the earliest assessment occurring on average at 11:50 AM (SD=2.6), the latest assessment occurring on average at 7:43 PM (SD=2.1). The average time-lapse between the morning assessment and the earliest assessment used to calculate the later-day composite score was 3.6 hours (SD=2.3).

Data preparation steps included separating time-varying predictor variables into the between-person and within-person components. The time-invariant between-person component, MorningAffectBP_i, was calculated as the intraindividual mean across the repeated measures of positive affect and separately for negative affect, yielding one score per person per affect valence. Similarly, the momentary within-person component, MorningAffectWP_it was calculated for each person's observation and affect valence (positive, negative) as the deviation from their intraindividual mean (MorningAffectBP_i). For models with smoking urge as the outcome, we included the within-person later-day smoking variable (0=no later day smoking, 1=yes, later-day smoking occurred) as a covariate. Additionally, we included as a covariate the between-person component for later-day smoking, calculated as the proportion of days during the post-quit study period with any later-day smoking behavior. For models with smoking behavior as the outcome, we also included as covariates the within- and between-person components for later-day smoking urges. Specifically, we calculated the between-person component of this variable as the intraindividual mean across the repeated measures of later-day smoking urges, yielding one score per person. Similarly, the momentary within-person component was calculated for each repeated measure for each person as the deviation from their intraindividual mean. Prior to analysis, the between-person components for affect and continuous scale covariates (i.e., baseline smoking dependence, the proportion of days with smoking behavior, characteristic later day smoking urges, and age) were sample-mean centered to facilitate interpretation regarding the prototypical person in the sample.

Statistical Models

To examine (1) the extent of within-person association between morning affect and later day smoking urges, and (2) the extent of between-person association between morning affect and later day smoking urges, we fit a multilevel linear regression model,

$${LaterDaySmokingUrge}_{it}={\beta }_{0i}+{\beta }_{1i}({MorningAffectWP}_{it})+{\epsilon }_{it}$$ (1)

where repeated measures of later-day smoking urges for participant $i$ during day $t$ are modeled as a function of a person-specific intercept, ${\beta }_{0i}$, changes driven by morning affect, ${\beta }_{1i}$, and residual differences, ${\epsilon }_{it}$. Person-specific coefficients were simultaneously modeled as a function of person-level predictors,

$${\beta }_{0i}={\gamma }_{00}+{\gamma }_{01}({MorningAffectBP}_i)+u_{0i}$$ (2)

$${\beta }_{1i}={\gamma }_{10}+u_{1i}$$ (3)

where ${\gamma }_{00}$, and ${\gamma }_{10}$ are sample-level parameters describing the prototypical person, and ${\gamma }_{01}$ describes how individual differences in morning affect are associated with participants’ later day smoking urges. Random effects ($u_{0i}$, $u_{1i}$) were allowed to co-vary with one another but not with ${\epsilon }_{it}$. Additionally, we examined whether the results held after accounting for intervention type, smoking factors (baseline smoking dependence and daily smoking behavior), days since quit date, and demographic variables (gender, race, age, and education).

To examine (1) the extent of within-person association between morning affect and the odds of smoking later that day and (2) the extent of between-person association between morning affect and the odds of later-day smoking, we fit a multilevel logistic regression model,

$$logit\left(\frac{p({LaterDaySmoking}_{it})}{1-p({LaterDaySmoking}_{it})}\right)={\beta }_{0i}+{\beta }_{1i}({MorningAffectWP}_{it})$$ (4)

where repeated measures of later-day smoking for participant $i$ on day $t$ were modeled as a function of a person-specific intercept, ${\beta }_{0i}$; and daily changes in morning affect. Person-specific coefficients were simultaneously modeled as a function of person-level predictors,

$${\beta }_{0i}={\gamma }_{00}+{\gamma }_{01}({MorningAffectBP}_i)+u_{0i}$$ (5)

$${\beta }_{1i}={\gamma }_{10}+u_{1i}$$ (6)

where ${\gamma }_{00}$, and ${\gamma }_{10}$, and are sample-level parameters describing the prototypical person, and ${\gamma }_{01}$ describes how individual differences in morning affect are associated with the odds of smoking later in the day. Random effects ($u_{0i}$, $u_{1i}$) were allowed to co-vary.

We also examined whether the results differed conceptually after accounting for intervention type, smoking factors (baseline smoking dependence, proportion of days with smoking behavior), days since quit date, and demographic variables (gender, race, age, and education). For models with smoking urge as the outcome, we included the momentary later-day smoking variable described above as a covariate. Additionally, we calculated the between-person component for the later-day smoking variable as the proportion of days during the post-quit study period with any later-day smoking behavior. For models with smoking behavior as the outcome, we included the later-day smoking urge variable, described above, as a covariate. Specifically, we calculated the between-person component of this variable as the intraindividual mean across the repeated measures of later-day smoking urges, yielding one score per person.

As a point of clarification, we did not include morning smoking urges as a predictor in models with later-day smoking urges as the outcome, nor did we include morning smoking as a predictor in models with later-day smoking as the outcome. Had we included those terms, the estimates and conceptual interpretation of the results would have differed. For example, instead of examining the within-person association between morning affect and later-day smoking urges, we would instead have examined the within-person association between morning affect and change in smoking urges from morning to later-day.

All analyses were performed in R version 4.1.1. and R Studio version 2021.09.0 using the brms package (R Core Team, 2020; RStudio Team, 2021; Bürkner, 2018). Pragmatically, models were fit in the Bayesian framework because the number and structure of random effects included make these models too complex to be estimated using frequentist optimization. We made this choice given the conceptual importance of random effects (i.e., allowing participants to differ in the extent of within-person coupling between their affective experiences and later-day smoking urges/ smoking). All figures were created in R using the ggplot2 package (Wickham, 2016). Models were fit in the Bayesian statistical framework to facilitate model convergence with full random effects, given that there were convergence issues when the models were fit in the frequentist framework. For each model, we made use of default (non-informative) prior options in the brms package. Specifically, the default prior for the fixed effects was a uniform distribution from negative infinity to positive infinity. The priors for the random effects were restricted to be non-negative, and were set to have a half student-t prior with 3 degrees of freedom and a scale parameter of 2.5 for each model. For the correlations among the random effects, brms set a Cholesky LKJ correlation distribution with eta equal to 1, indicating that all correlation matrices were equally likely a prior. We also specified two chains with 7,000 samples each (warm-up=1,000 samples, post-warm-up samples=6,000), and set thin equal to 1, meaning that we did not discard any samples. Altogether, we attained 12,000 samples from the posterior distribution. Inferences were guided by the posterior mean and 95% credible interval (CI) of each parameter. The 95% CI marks the proportion of the posterior distribution that contains 95% of the posterior sample. Given the observed data, there is a 95% probability that the true parameter is contained in the 95% CI. Therefore, we consider a posterior mean estimate to be indicative of a meaningful association if its 95% CI does not span zero.

Results

In the morning, participants tended to experience relatively high positive affect (M=3.6, SD=0.7) and relatively low negative affect (M=2.0, SD=0.8) during post-quit days when they did not smoke during or prior to the morning assessment. Prior to the main analyses, we calculated intraclass correlation coefficients (ICC). For positive affect, 38% of the variation could be considered within-person variation across days, with the remaining 62% considered as between-person variation. For negative affect, 27% of the variation was within-person and 74% of the variation was between-persons. Smoking urges later in the day were on average, rated as neutral (M=2.7 SD=0.9). Additionally, on average, participants smoked later in the day for 31% of the post-quit study days (SD=35, Min=0%, Max=100%). The ICC for later-day smoking urges indicated 32% of the variation could be considered within-person variation across days, and 68% considered between-person variation. For later-day smoking, 46% of the variation was within-person and 54% was between-persons. Given considerable variation at both levels of analysis, we proceeded to fit the multilevel models described above. Results are reported in the following sections. Exact analysis code and model output are available on OSF (AnalysisCode_AffectSmoking_OSF.Rmd located at https://osf.io/78ngp/). This file also contains information on model convergence. In brief, for all parameters in the model, the $\widehat{R}$ values were well below the heuristic of 1.1, providing indication that there were not problems with model convergence and evidence that the two chains were adequate. Similarly, inspection of the trace plots indicated that the two chains mixed properly for each parameter in each model.

Associations between Morning Negative Affect and Later Day Smoking Urges

First, we examined whether between-person differences and within-person variation across days in morning negative affect were associated with later-day smoking urges. As shown in Table 2, results indicated that, for a prototypical participant in the sample with average characteristic morning negative affect, on a day when their negative affect in the morning was at their usual level, the intensity of later-day smoking urges was expected to be 2.7 (max=5; ${\gamma }_{00}$). The results also showed credible between- and within-person associations between morning negative affect and later day smoking urges. As shown in the top right panel of Figure 1, individuals who, on average, experienced higher negative affect in the morning also tended to experience more intense smoking urges later in the day (${\gamma }_{01}=0.7$). Similarly, as shown in the top left panel of Figure 1, on mornings when individuals’ negative affect was higher than usual, their smoking urges later in the day also tended to be more intense than usual (${\gamma }_{10}=0.3$). The results also showed that participants differed substantially from one another in the extent of within-person association between morning negative affect and later day smoking urges, indicated by the light gray lines in the top left panel of Figure 1 (${\sigma }_{u1}=0.4$).

Table 2.

Results from Multilevel Models Examining Associations among Negative Affect, Smoking Urges and Smoking Behavior

		Smoking Urges Linear Model				Smoking Behavior Logistic Model
		Primary Model		Covariate Adjusted Model		Primary Model		Covariate Adjusted Model
	Parameter	Est.	95% CI	Est.	95% CI	Est.	95% CI	Est.	95% CI
Fixed Effects
Intercept	${\gamma }_{00}$	2.7	[2.5, 2.9]	2.7	[2.2, 3.2]	−1.8	[−2.8, −0.9]	−1.6	[−4.4, 1.3]
MorningNegAffectWP	${\gamma }_{10}$	0.3	[0.1, 0.4]	0.2	[0.1, 0.3]	0.7	[0.1, 1.2]	0.5	[−0.1, 1.1]
MorningNegAffectBP	${\gamma }_{01}$	0.7	[0.5, 0.9]	0.7	[0.5, 0.9]	0.4	[−0.8, 1.6]	−1.5	[−3.2, 0.2]
TxQG	${\gamma }_{02}$	-	-	−0.3	[−0.8, 0.2]	-	-	−0.1	[−2.9, 2.8]
TxTTRP	${\gamma }_{03}$	-	-	−0.2	[−0.6, 0.3]	-	-	−0.9	[−3.7, 1.6]
BaselineHSI	${\gamma }_{04}$	-	-	0.1	[−0.1, 0.2]	-	-	0.7	[−0.2, 1.7]
LaterDaySmokingBP*	${\gamma }_{05}$	-	-	0.57	[0.03, 1.10]	-	-	2.2	[0.7, 3.8]
White	${\gamma }_{06}$	-	-	0.01	[−0.37, 0.39]	-	-	−0.9	[−3.1, 1.3]
Female	${\gamma }_{07}$	-	-	0.2	[−0.1, 0.6]	-	-	0.6	[−1.4, 2.6]
Age	${\gamma }_{08}$	-	-	−0.01	[−0.02, 0.01]	-	-	−0.02	[−0.12, 0.07]
College	${\gamma }_{09}$	-	-	0.38	[−0.02, 0.77]	-	-	−0.1	[−2.6, 2.3]
LaterDaySmokingWP*	${\gamma }_{20}$	-	-	0.12	[0.01, 0.25]	-	-	0.5	[0.1, 0.9]
DaysSinceQuitDate	${\gamma }_{30}$	-	-	−0.02	[−0.03, −0.01]	-	-	0.01	[−0.04, 0.05]
Random Effects
SD Intercept	${\sigma }_{u0i}$	0.8	[0.6, 0.9]	0.7	[0.5, 0.8]	3.4	[2.6, 4.5]	3.5	[2.4, 4.9]
SD MorningNegAffectWP	${\sigma }_{u1i}$	0.4	[0.3, 0.5]	0.3	[0.2, 0.4]	0.66	[0.03, 1.63]	0.44	[0.02, 1.25]
SD LaterDaySmokingWP	${\sigma }_{u2i}$	-	-	0.19	[0.02, 0.36]	-	-	0.48	[0.03, 1.09]
SD DaysSinceQuitDate	${\sigma }_{u3i}$	-	-	0.04	[0.03, 0.05]	-	-	0.10	[0.05, 0.16]
Correlation Intercept, MorningNegAffectWP	$\rho [u_{0i},u_{1i}]$	−0.3	[−0.6, 0.1]	−0.2	[−0.6, 0.3]	0.3	[−0.8, 1.0]	−0.1	[−0.8, 0.8]
Correlation Intercept, LaterDaySmokingWP	$\rho [u_{0i},u_{2i}]$	-	-	0.1	[−0.6, 0.7]	-	-	−0.1	[−0.8, 0.7]
Correlation Intercept, DaysSinceQuitDate	$\rho [u_{0i},u_{3i}]$	-	-	−0.27	[−0.53, 0.03]	-	-	−0.1	[−0.6, 0.5]
Correlation MorningNegAffectWP, LaterDaySmokingWP	$\rho [u_{1i},u_{2i}]$	-	-	0.4	[−0.5, 0.9]	-	-	0.1	[−0.8, 0.8]
Correlation MorningNegAffectWP, DaysSinceQuitDate	$\rho [u_{1i},u_{3i}]$	-	-	−0.02	[−0.45, 0.42]	-	-	0.2	[−0.7, 0.9]
Correlation LaterDaySmokingWP, DaysSinceQuitDate	$\rho [u_{2i},u_{3i}]$	-	-	−0.1	[−0.7, 0.5]	-	-	0.1	[−0.7, 0.8]
Residual SD	${\sigma }_{\epsilon i}$	0.59	[0.57, 0.61]	0.51	[0.49, 0.53]	-	-	-	-

Notes. Est.=Posterior Mean; 95% CI=95% Credible Interval; BP=between-person; WP=within-person. Estimates are bolded if their corresponding 95% CI does not contain zero, indicating a credible association. Data are reports on 1,325 post-quit days nested within 63 people. In the model with smoking urge as the outcome, LaterDaySmokingBP* and LaterDaySmokingWP* pertain to smoking behavior whereas for the model with smoking behavior as the outcome, LaterDaySmokingBP* and LaterDaySmokingWP* pertain to smoking urges. Logistic model results are reported in log odds units.

Figure 1.

Within-person (left column) and between-person (right column) associations between morning negative affect and later-day smoking urges (top row), and later-day smoking behavior (bottom row). Light gray lines (left column) show the range of individual differences in the extent of the within-person associations. Results indicated evidence of credible associations for the top left panel, top right panel and bottom left panel.

In the model with covariates, the between- and within-person associations for morning negative affect and later-day smoking urges remained credible. Proportion of later-day smoking days (${\gamma }_{05}=0.57$), later-day smoking within a given day (${\gamma }_{20}=0.12$), and days since quit date (${\gamma }_{30}=-0.02$) were credibly associated with later-day smoking urges. There were also substantial differences across participants in the extent to which their later-day smoking related to their later-day smoking urges (${\sigma }_{u2}=0.19$).

Associations between Morning Negative Affect and Later Day Smoking Behavior

Second, we examined whether between-person differences and within-person variation across days in morning negative affect were associated with smoking later in the day. As shown in Table 2, for the prototypical participant in the sample with average characteristic morning negative affect, on a day with characteristic negative affect in the morning, their log odds of smoking at some point later in the day were −1.8 (${\gamma }_{00}$), indicating that participants were less likely to smoke than not smoke on any given day. Re-cast in probability space, on days when participants’ negative affect was at their usual or characteristic level, their probability of smoking that day was expected to be 0.15. In contrast, there was not a credible between-person association for morning negative affect and later-day smoking behavior (${\gamma }_{01}$). This indicates that those who tended to experience higher average morning negative affect did not tend to have higher or lower average log odds of smoking than those who tended to experience lower average negative affect in the morning. In contrast, as shown in the bottom left panel of Figure 1, there was a credible within-person association between morning negative affect and later-day smoking (${\gamma }_{10}=0.7$). Re-cast in the probability space, on days when participants’ negative affect was higher than usual (+1 SD), their probability of smoking that day was expected to be 0.19, whereas on days when their negative affect was lower than usual (−1 SD), their probability of smoking that day was expected to be 0.12. The results also showed substantial differences across participants in the extent of within-person association between morning negative affect and later-day smoking, indicating relatively stronger within-person coupling for some individuals and relatively weaker within-person coupling for other participants (${\sigma }_{u1}=0.7$). This is visually depicted by the faint gray lines in the bottom left panel of Figure 1.

In the model adjusting for covariates, the within-person association between morning negative affect and later-day smoking was no longer credible (${\gamma }_{10}=0.5$, 95% CI[−0.1, 1.1]). Of all covariates, only between-person differences in later-day smoking urges (${\gamma }_{05}=2.2$), and within-person variation in later-day smoking urges (${\gamma }_{30}=0.5$) were associated with the log odds of later-day smoking. There were also substantial differences across participants in the extent to which their later-day smoking urges were associated with their odds of later-day smoking (${\sigma }_{u2}=0.5$).

Associations between Morning Positive Affect and Later Day Smoking Urges

Third, we examined whether between-person differences and within-person variation across days in morning positive affect were associated with smoking urges later in the day. As shown in Table 3, the results indicated that for the prototypical participant in the sample with average characteristic morning positive affect, on a day when their positive affect in the morning was at their usual level, the intensity of later-day smoking urges was expected to be 2.7 (max=5; ${\gamma }_{00}$). Results indicated credible between- and within-person associations between morning positive affect and later day smoking urges. As shown in the top right panel of Figure 2, individuals who, on average, experienced higher positive affect in the morning also tended to experience less intense smoking urges later in the day (${\gamma }_{01}=-0.7$). Similarly, as shown in the top left panel of Figure 2, on mornings when individuals’ positive affect was higher than usual, their smoking urges later in the day tended to be less intense than usual (${\gamma }_{10}=-0.1$). The results also showed that participants differed substantially from one another in the extent of within-person association between morning positive affect and later day smoking urges, indicated by the light gray lines in the top left panel of Figure 2 (${\sigma }_{u1}=0.2$).

Table 3.

Results from Multilevel Models Examining Associations among Positive Affect, Smoking Urges and Smoking Behavior

		Smoking Urges Linear Model				Smoking Behavior Logistic Model
		Primary Model		Covariate Adjusted Model		Primary Model		Covariate Adjusted Model
	Parameter	Est.	95% CI	Est.	95% CI	Est.	95% CI	Est.	95% CI
Fixed Effects
Intercept	${\gamma }_{00}$	2.7	[2.5, 2.9]	2.7	[2.1, 3.2]	−1.8	[−2.8, −0.9]	−1.5	[−4.3, 1.4]
MorningPosAffectWP	${\gamma }_{10}$	−0.12	[−0.22, −0.02]	−0.12	[−0.20, −0.04]	−0.2	[−0.7, 0.3]	−0.2	[−0.7, 0.2]
MorningPosAffectBP	${\gamma }_{01}$	−0.7	[−1.0, −0.4]	−0.7	[−1.0, −0.4]	−0.3	[−1.6, 1.0]	1.0	[−0.9, 2.8]
TxQG	${\gamma }_{02}$	-	-	−0.1	[−0.5, 0.5]	-	-	−0.6	[−3.5, 2.1]
TxTTRP	${\gamma }_{03}$	-	-	−0.1	[−0.6, 0.3]	-	-	−1.0	[−3.7, 1.6]
BaselineHSI	${\gamma }_{04}$	-	-	0.20	[0.04, 0.36]	-	-	0.6	[−0.4, 1.6]
LaterDaySmokingBP*	${\gamma }_{05}$	-	-	0.52	[−0.06, 1.09]	-	-	1.9	[0.4, 3.4]
White	${\gamma }_{06}$	-	-	−0.1	[−0.5, 0.3]	-	-	−0.6	[−2.8, 1.5]
Female	${\gamma }_{07}$	-	-	0.2	[−0.2, 0.6]	-	-	0.6	[−1.3, 2.7]
Age	${\gamma }_{08}$	-	-	−0.01	[−0.02, 0.01]	-	-	−0.02	[−0.11, 0.07]
College	${\gamma }_{09}$	-	-	0.42	[−0.02, 0.86]	-	-	−0.01	[−2.6, 2.4]
LaterDaySmokingWP*	${\gamma }_{20}$	-	-	0.14	[0.02, 0.27]	-	-	0.5	[0.1, 0.9]
DaysSinceQuitDate	${\gamma }_{30}$	-	-	−0.02	[−0.03, −0.01]	-	-	0.01	[−0.04, 0.05]
Random Effects
SD Intercept	${\sigma }_{u0i}$	0.8	[0.7, 1.0]	0.7	[0.6, 0.9]	3.4	[2.6, 4.4]	3.5	[2.4, 5.0]
SD MorningPosAffectWP	${\sigma }_{u1i}$	0.2	[0.1, 0.4]	0.15	[0.04, 0.26]	0.7	[0.1, 1.4]	0.42	[0.02, 1.13]
SD LaterDaySmokingWP	${\sigma }_{u2i}$	-	-	0.21	[0.04, 0.38]	-	-	0.47	[0.03, 1.09]
SD DaysSinceQuitDate	${\sigma }_{u3i}$	-	-	0.04	[0.03, 0.05]	-	-	0.10	[0.06, 0.16]
Correlation Intercept, MorningPosAffectWP	$\rho [u_{0i},u_{1i}]$	−0.1	[−0.5, 0.3]	−0.1	[−0.7, 0.4]	−0.4	[−1.0, 0.6]	0.2	[−0.7, 0.9]
Correlation Intercept, LaterDaySmokingWP	$\rho [u_{0i},u_{2i}]$	-	-	0.1	[−0.6, 0.7]	-	-	−0.1	[−0.8, 0.7]
Correlation Intercept, DaysSinceQuitDate	$\rho [u_{0i},u_{3i}]$	-	-	−0.31	[−0.56, −0.02]	-	-	−0.1	[−0.6, 0.5]
Correlation MorningPosAffectWP, LaterDaySmokingWP	$\rho [u_{1i},u_{2i}]$	-	-	−0.5	[−0.9, 0.3]	-	-	0.01	[−0.80, 0.82]
Correlation MorningPosAffectWP, DaysSinceQuitDate	$\rho [u_{1i},u_{3i}]$	-	-	0.5	[−0.1, 0.8]	-	-	−0.2	[−0.9, 0.6]
Correlation LaterDaySmokingWP, DaysSinceQuitDate	$\rho [u_{2i},u_{3i}]$	-	-	−0.2	[−0.7, 0.4]	-	-	0.1	[−0.7, 0.8]
Residual SD	${\sigma }_{\epsilon i}$	0.61	[0.59, 0.63]	0.52	[0.50, 0.54]	-	-	-	-

Notes. Est.=Posterior Mean; 95% CI=95% Credible Interval; BP=between-person; WP=within-person. Estimates are bolded if their corresponding 95% CI does not contain zero, indicating a credible association. Data are reports on 1,325 post-quit days nested within 63 people. In the model with smoking urge as the outcome, LaterDaySmokingBP* and LaterDaySmokingWP* pertain to smoking behavior whereas for the model with smoking behavior as the outcome, LaterDaySmokingBP* and LaterDaySmokingWP* pertain to smoking urges. Logistic model results are reported in log odds units.

Figure 2.

Within-person (left column) and between-person (right column) associations between morning positive affect and later-day smoking behavior (top row), and between morning negative affect and later-day smoking behavior (bottom row). Light gray lines (left column) show the range of individual differences in the extent of the within-person associations. Results indicated evidence of credible associations for the top left panel and top right panel.

In the model with covariates, both between- and within-person associations for morning positive affect and later-day smoking urges remained credible. Three covariates were credibly associated with later-day smoking urges including baseline smoking dependence (${\gamma }_{04}=0.2$), later-day smoking (${\gamma }_{20}=0.14$), and days since quit date (${\gamma }_{30}=-0.02$). There were also substantial differences across participants in the extent to which their later-day smoking behavior related to their later-day smoking urges (${\sigma }_{u2}=0.2$). This suggests that some participants experienced relatively strong positive coupling between their smoking urges and smoking behavior, whereas others experienced relatively weak coupling between their smoking urges and smoking.

Associations between Morning Positive Affect and Later Day Smoking Behavior

Fourth, we examined whether between-person differences and within-person variation across days in morning positive affect were associated with smoking cigarettes later in the day. As shown in Table 3, the results indicated that, for a prototypical participant in the sample with average characteristic morning positive affect, on a day with characteristic positive affect in the morning, their log odds of smoking at some point later in the day were −1.8 (${\gamma }_{00}$), indicating that participants were less likely to smoke than not smoke on any given day. There was not a credible between-person association for morning positive affect and later-day smoking (${\gamma }_{01}$). This indicates that those who tended to have higher characteristic morning positive affect did not have higher or lower average log odds of smoking than those who tended to have relatively lower characteristic morning positive affect. There was also not a credible within-person association between morning positive affect and later-day smoking (${\gamma }_{10}$). This indicates that on mornings when individuals’ positive affect is higher than usual, their odds of smoking later in the day do not systematically differ from mornings when their positive affect is lower than usual. However, the results indicated substantial differences across participants in the extent of within-person association between morning positive affect and later day smoking, indicating stronger coupling for some individuals and weaker coupling for others (${\sigma }_{u1}=0.7$). This is visually depicted by the faint gray lines in the top left panel of Figure 2.

In the model with covariates, the results indicated neither a within- nor a between-person association between morning positive affect and later-day smoking. Of all covariates, between-person differences in later-day smoking urges (${\gamma }_{05}=1.9$), and within-person variation in later-day smoking urges (${\gamma }_{30}=0.5$) were associated with the log odds of later-day smoking behavior. There were also substantial differences across participants in the extent to which their later-day smoking urges were associated with their odds of later-day smoking behavior (${\sigma }_{u2}=0.47$).

Discussion

The present study is the first to use EMA data to examine within- and between-person associations between morning affective experiences and later-day smoking urges and smoking behaviors in adults who were actively trying to quit smoking while receiving nicotine replacement therapy and either counseling-based or app-based intervention support. The results showed that positive affect in the morning was associated with later-day smoking urges at both the within- and between-person levels of analysis, but not associated with later-day smoking behavior at either the within- or between-person levels of analysis. Additionally, the results showed that negative affect in the morning was associated with later-day smoking urges at both the within- and between-person levels of analysis and also associated with later-day smoking behavior at the within-person level but not between-person level. Finally, the results showed that there were substantial between-person differences in the extent of within-person association between morning positive and negative affective experiences, and later-day smoking urges and smoking behavior. Altogether, results from the present study contribute to a growing literature on the affective determinants of smoking experiences in the context of smoking cessation and elucidate a specific within-day time-scale that could be targeted in future intervention development.

Morning Negative Affect and Later-Day Smoking Urges and Smoking Behavior

In line with expectations from the affective processing model of negative reinforcement (Baker et al., 2004), results from the present study suggest that moments of heightened negative affect were associated with subsequent smoking urges and smoking behavior. Specifically, at the within-person level, the results showed that on mornings when individuals’ negative affect was higher than their usual, their smoking urges later in the day also tended to be higher than usual. Prior research in this domain has operationalized negative affect as negative mood (assessed via a bipolar scale ranging from extremely positive to extremely negative) and anxious mood (assessed via a bipolar scale ranging from calm to anxious), and found evidence of prospective associations with subsequent within-day smoking urges (Shiyko et al., 2013; Serre et al., 2018). The present study aligns with and builds on this prior research by showing that there are also prospective within-day associations between negative affect and subsequent smoking urges when negative affect is operationalized as the composite of several discrete negative emotions (e.g., anxious, angry, sad) assessed via a unipolar measurement scale (i.e., Strongly Disagree to Strongly Agree). In complement to the within-person findings, at the between-person level of analysis, we found that that those who tended to have higher characteristic morning negative affect also tended to have higher smoking urges later in the day. Altogether, these findings suggest that morning/later-day is a relevant time-scale for capturing prospective associations between negative affect and smoking urges. These findings also imply that future smoking cessation interventions may benefit from targeting both global negative affect during the first several weeks after a person’s quit date as well as daily variation in morning negative affect.

Results from the present study also indicate that on mornings when individuals’ negative affect was higher than usual, their odds of later-day smoking were also higher than usual. Similar to the above results for smoking urges, this finding aligns with several prior studies that operationalized negative affect based on both bipolar scales (e.g., Berkman et al., 2011; Shiffmann & Waters, 2004) and composites of negative valence emotions assessed via unidimensional scales (Moore et al., 2014; Potter et al., 2021). This finding also builds on prior research by demonstrating the prospective within-person association holds when operationalizing time as morning and later-day rather than time lags occurring at any time throughout the day. It should be noted however, that once covariates were introduced into the model, the within-person association between morning negative affect and likelihood of later-day smoking was no longer credible. Although the posterior mean estimates of both models were relatively similar and in the same direction, this finding should be interpreted with caution and future research will be needed to examine whether this finding replicates.

Interestingly, results from the present study did not provide evidence of a between-person association between characteristic morning negative affect and later-day smoking tendencies. This finding motivates the need for examining negative affective experiences and smoking behaviors intensively during a quit attempt. This finding also suggests that even if negative affect is heightened overall during a quit attempt, interventions can still support individuals in their smoking cessation goals by catching them on mornings when their negative affect is unusually high and providing just-in-time aid. The Smart-T2 intervention provided to one group of participants in the present study is an example of a just-in-time intervention that includes tailored support based on the most highly endorsed risk factor (Hébert et al., 2018). Future refinement of this automated intervention or other interventions may benefit from monitoring morning affect and automatically intervening when an individual’s affect is higher or lower than their usual.

Morning Positive Affect and Later-Day Smoking Urges and Smoking Behavior

The present study is the first to examine prospective within-day associations between morning positive affect and later-day smoking urges and the results generally aligned with expectations based on the dynamic model of affect (Zautra et al., 2001). At the within-person level, the results showed that on mornings when individuals’ positive affect was higher than usual, their later-day smoking urges tended to be lower than usual. This suggests that interventions that boost positive affect may facilitate individuals’ resistance to smoking urges. Similarly, at the between-person level, we found that those who tended to experience higher positive affect in the morning on average during the post-quit period also tended to experience lower later-day smoking urges. This additionally suggests that regularly engaging in activities that help maintain relatively high overall levels of positive affect during the initial weeks of a smoking cessation attempt may be beneficial. In line with this idea, prior research suggests that positive psychology-based interventions may be beneficial for both positive affect and smoking cessation (Kahler et al., 2015; Kahler et al., 2018). Future research would benefit from examining whether just-in-time adaptive interventions designed to boost positive affect can lower the intensity of smoking urges and reduce the likelihood of smoking later in the day when individuals report lower positive affect than their usual in the morning.

Contrary to expectation based on theory and previous research, we did not find evidence of within- or between-person associations between morning positive affect and later-day smoking behavior. Past research motivates two possible explanations for these unexpected findings. First, it is possible that a prospective within-day association between positive affect and subsequent smoking urges only occurs for a particular time frame in the post-quit period. Specifically, Potter et al. (2021) found the prospective within-day association between positive affect and odds of subsequent smoking were only significant for days three through 11 post-quit. Second, it may also be that the prospective within-person association between positive affect and subsequent smoking behavior occurs on a slower time-scale across days rather than within a day (Ferguson et al., 2006). Altogether, given the relatively low number of past studies that have examined this association, future research is needed to determine whether and over what time-scale(s) a prospective association exists between positive affect and smoking behavior.

An alternative explanation to the time-scale considerations discussed above is that the prospective within-person association between positive affect and smoking behavior differs based on between-person characteristics. While results from the present study did not indicate there was a credible average within-person association between positive affect and later-day smoking behavior, there were substantial between-person differences in the extent of coupling. The credible random effect suggested that for some individuals there was positive coupling between morning positive affect and later day smoking behavior, whereas for other individuals there was negative coupling or no coupling. Thus, future work would benefit from examining this association in a larger sample of individuals to facilitate exploration of moderators of this within-person association. For example, the extent of smoking during the post-quit period may be an important moderator to consider. In the present study, participants were included in the analyses regardless of how many post-quit days they smoked, and were only removed from the analyses if they smoked at or prior to the morning occasion for all post-quit assessments. In contrast, other approaches exclude participants who smoked for seven or more consecutive days during the post-quit period (Potter et al., 2021). Furthermore, given that participants differed substantially from one another in the strength of the association between their morning positive affect and their later-day smoking behavior, future research would benefit from exploring whether different patterns of abstinence can help explain such between-person differences in this within-person association. Finally, this finding aligns with prior research suggesting individuals differ in terms of which psychological states and/or contextual factors may be related to engaging in health behaviors, suggesting the need to fit person-specific idiographic models to better understand and predict smoking behavior (Molenaar, 2004).

Strength, Limitations, and Future Directions

This study has several strengths, limitations, and future directions. A strength is that data were collected using an intensive monitoring design, including up to 5 EMAs per day for the first 4 weeks of a smoking cessation attempt and additional information gathered from participant-initiated assessments. This afforded opportunities to examine within-person associations among affective experiences and smoking experiences during the process of smoking cessation, using ecologically valid data collected from individuals in their daily lives. In contrast, as has been reported elsewhere (e.g., Benson et al., 2022), data for the present study were part of a pilot study with a relatively small number of participants who were primarily White, non-Hispanic, and recruited from a specific region of the United States, thus limiting generalizability of the present study findings. The analytic sample size included 63 people, with approximately 21 participants per intervention group. Due to the relatively small sample size, the between-person findings for morning affect and later-day smoking urges and smoking should be interpreted with caution. Based on results from a simulation study we conducted, power considerations may be particularly important when there are substantial between-person differences in the random intercept. See the Appendix for more details that may be useful for future studies involving new data collection. Given that the present study research questions were exploratory, future research is needed to better understand the extent to which these findings replicate, how these results generalize to different demographics, and whether intervention types moderate the within-person associations. Future work in this domain involving new data collection would also benefit from a priori power calculations.

Another data consideration pertains to the methods used to collect the smoking and affect data. The present study made use of the daily self-report data. A strength of this approach is that intensive monitoring is possible given that it only takes a few minutes to answer questions for each assessment, thus minimizing participant burden and minimizing the potential for recall bias (Shiffman et al., 1997). One limitation of the decision to limit participant burden is that relatively few discrete positive emotions were assessed (i.e., calm, happy). This limited measurement scheme may have missed several types of positive emotion experiences relevant to smoking cessation. In contrast, prior research has shown that ‘calm’ was one of the strongest predictors of imminent smoking lapse (Suchting et al., 2019). Future research would benefit from further study of positive emotion experiences during the smoking cessation process.

A strength is that by using both the random assessments and participant-initiated assessments, the estimate of whether someone smoked on a given day has the potential to be more accurate since smoking behavior occurring after the last random assessment for the day could still be captured. There are also limitations, however, such that the self-report nature of the assessments may be biased due to factors such as social desirability (Murray et al., 1993), and some participants may be less compliant than others in responding to the random assessments and/or self-initiating reports of their smoking behavior in the moment. Similarly, while the 5 daily prompted assessments and self-initiated reports of smoking provided a detailed glimpse into individuals’ daily smoking experiences, it is also possible that some smoking episodes were also missed by the measurement design. Additionally, self-reported smoking status was biochemically verified in this study, but these assessments only occurred during the in-person assessments (i.e., quit date and four weeks post-quit), and thus did not match the much faster within-day time-scale examined in the present study. As sensor-based technology for smoking detection continues to advance, future research should consider less burdensome passive and continuous sensing of smoking behavior to avoid biases due to self-report (Saleheen et al., 2015).

While the results from the present study indicate morning affect experiences may be useful intervention targets for future smoking cessation interventions, a limitation of the present study is that the results do not apply to days when participants had smoked during or before the morning assessment. This limits generalizability of the present study findings given that early morning smoking, including waking up to smoke in the middle of the night or smoking within minutes of waking up, are markers of smoking dependence (Fagerstrom, 2003). In the present study, at least some data were excluded from nearly two thirds of the participants (271 days across all participants) during the data preparation stage due to participants smoking on or before the morning assessment. Given the smoking risk associated with the early morning timeframe, future research may benefit from examining prospective cross-day within-person associations between evening affective experiences and early morning smoking urges and behaviors.

In addition to data-related strengths and limitations, the present study also has strengths and limitations pertaining to the modeling approach. Specifically, the present study considered the first four weeks of smoking cessation as a single period of time and used a Bayesian multilevel model to examine within-person associations between morning affective experiences and later-day smoking experiences and the extent of between-person differences in these within-person associations. It may also be that slower time-scale dynamics (e.g., adaptation to the intervention) were happening in tandem with the faster within-day time-scale dynamics (Ram et al., 2014). Future work may benefit from using a time-varying parameter model to examine whether the within-person association between morning affective experiences and later-day smoking experiences changes over the first four weeks of a quit attempt (Haslbeck et al., 2020).

Public Health Significance Statement:

This study indicates that individuals who experience higher than usual negative affect on a given morning may be at heightened risk for smoking urges and smoking behavior later in the day whereas on mornings when individuals experience higher than usual positive affect, their risk for smoking urges later in the day may be reduced. These finding suggest that even if negative affect is heightened overall during a quit attempt, interventions might still support individuals in their smoking cessation goals by catching them on mornings when their negative affect is unusually high and providing just-in-time aid.

Synopsis.

Overall, the present study findings suggest that on mornings when individuals’ positive affect is higher than usual, their later-day smoking urges (but not smoking behavior) tend to be lower than usual whereas on mornings when individuals’ negative affect is higher than usual, their later-day smoking urges and the likelihood of later-day smoking both tend to be higher than usual. We also observed substantial between-person differences in these within-person associations, which motivates the need for future research to identify moderators that might explain such between-person differences. The present study results also motivate the need for future research to examine whether mobile health interventions designed to boost positive affect on mornings when it is lower than usual, and attenuate negative affect on mornings when it is higher than usual, can reduce smoking urges and the likelihood of smoking later in the day. Finally, the present study extends existing research by demonstrating the relevance of morning and later-day as a socially constructed time-scale for examining prospective associations between affective experiences and smoking experiences during the smoking cessation process, which can be further evaluated within the context of future smoking cessation interventions.

Appendix

Power Simulation for Detecting a Range of Effect Sizes

We conducted a sample size analysis to determine the power of detecting a range of effect sizes for a between-person predictor and a within-person predictor in both logistic and linear multilevel regression models, under different parameter assumptions.

For the multilevel logistic regression model, we investigated the power under three conditions of random effect standard deviations (SDs): small (condition SD1, both random intercept and random slope SDs = 0.1), medium (condition SD2, both random intercept and random slope SDs = 0.4), and a mixed combination of two SDs (that is similar to what we observed in the results of the present study analyses; condition SD3, random intercept SD = 3.4, random slope SD = 0.3). With medium-and-above-sized SDs (SD2 and SD3), we also investigated the model performance under different correlations between the random effects (condition C1: r=0, C2: r=0.3, C3: r=0.5, C4: r=0.7). We did not include SD1 in the correlation design as we expected the correlation to have a small influence when the SDs are small. In addition, we included four different fixed effect intercept conditions (I1 = −1.8, I2 = −0.7, I3 = 0, I4 = 0.7, which roughly correspond to odd ratios of 0.17, 0.5, 1, and 2) and examined their influences on power in the contexts of SD1 (least likely to be affected by random effects) and SD3 (most likely to be affected by random effects). For each specific condition, we investigated standard effect sizes from 0.1 in increments of 0.1 and stopped when the observed power for both the between-person and within-person predictor reached 1 (100%).

In multilevel linear regression models, the intercept is less likely to influence the estimates of the between-person and within-person parameters. Therefore, we only varied the random effect SDs. Specifically, SD1 and SD2 are the same as in the multilevel logistic regression model described above. SD3’ was specified to have a random intercept SD of 0.7 and a random slope SD of 0.2. The correlation between the random intercept and random slope (C1-C4) are identical to those specified in the multilevel logistic regression model. We adopted the same sequence of standardized effect sizes and procedure as for the multilevel logistic regression model.

Results of the power simulation are reported in Table 1 for the multilevel logistic regression model and Table S2 for the multilevel linear regression model. As shown in Figure 1 and Figure 4, with all other parameters held identical, a higher random-effect SD resulted in a lower power for both the within- and between-person effects in both the linear and logistic models we investigated. In particular, the SD of the random intercept influences the power of detecting a between-person effect, as the between-person effect is on the same level of analysis as the random intercept. The fixed-effect intercept in the multilevel logistic regression model, which translates into average expected odds ratio, has an influence on both the power to observe a between-person effect and the power to observe a within-person effect. However, as shown in Figure 2, the influence of the fixed effect intercept was found to be more salient in the small random-effect SD condition. We speculate that having an average expected odds ratio closer to the border conditions (0 or 1) would result in diminished power to observe either effect in question. Finally, Figure 3 and Figure 5 show that the effect of varying the correlation between random effects on power was found to be minimal in both of the models.

Table 1.

Descriptives and Correlations among Main Study Variables

	Morning Positive Affect	Morning Negative Affect	Later-Day Smoking Urges	Later-Day Smoking Behavior
Morning Positive Affect (Range)	−	−0.8	−0.5	0.0
Morning Negative Affect (Range)	−0.4 (−1.0, 0.4)	-	0.6	0.1
Later-Day Smoking Urges (Range)	−0.1 (−0.7, 0.7)	0.2 (−0.5, 0.9)	-	0.4
Later-Day Smoking Behavior (Range)	−0.1 (−0.6, 0.5)	0.1 (−0.4, 0.7)	0.1 (−0.5, 0.6)	-
Sample Mean	3.6	2.0	2.7	0.3
Sample SD	0.7	0.5	0.9	0.4

Notes. Correlations in regular type indicate the extent of between-person association among person-level characteristics. Correlations in bold type indicate the average within-person correlation. Sample Mean is the sample average of the within-person means. Sample SD is the standard deviation of the sample average of the within-person means.

Figure A1.

Power under different random-effect SDs for the multilevel logistic regression model. Correlation between the random intercept and random slope is set as zero for conditions in these two plots.

Figure A2.

Power under different fixed-effect intercepts for the multilevel logistic regression model. The top two plots are with medium-to-large random effect SDs (SD3) and the bottom two with small random-effect SD (SD1).

Figure A3.

Power under different correlation between random effects for the multilevel logistic regression model. The top two plots are with medium-to-large random effect SDs (SD3) and the bottom two with medium random-effect SD (SD2).

Figure A4.

Power under different random-effect SDs for the multilevel linear regression model. Correlation between the random intercept and random slope is set as zero for conditions in these two plots.

Figure A5.

Power under different random-effect SDs for the multilevel linear regression model. Correlation between the random intercept and random slope is set as zero for conditions in these two plots.

Table A1.

Multilevel Logistic Regression Model

	Condition				Power for Specific Standardized Effect Sizes (Within-Person Effect, Between-Person Effect)
	Fixed Effect Intercept	Random Intercept SD	Random Slope SD Within- Person Predictor	Correlation between Random Intercept and Random Slope	0.1	0.2	0.3	0.4	0.5	0.6	0.7	0.8	0.9
SD1, C1, I1	−1.8	0.1	0.1	0	0.18, 0.27	0.65, 0.70	0.95, 0.94	1.00, 1.00	–	–	–	–	–
SD1,C1, I2	−0.7	0.1	0.1	0	0.35, 0.37	0.90, 0.92	1.00, 1.00	–	–	–	–	–	–
SD1, C1, I3	0	0.1	0.1	0	0.32, 0.42	0.86, 0.91	1.00, 0.98	1.00, 1.00	–	–	–	–	–
SD1, C1, I4	0.7	0.1	0.1	0	0.31, 0.27	0.88, 0.82	1.00, 0.98	1.00, 1.00	–	–	–	–	–
SD2, C1, I1	−1.8	0.4	0.4	0	0.16, 0.18	0.54, 0.59	0.85, 0.92	0.99, 0.98	1.00, 1.00	–	–	–	–
SD2, C2, I1	−1.8	0.4	0.4	0.3	0.24, 0.20	0.58, 0.62	0.90, 0.87	1.00, 0.98	1.00, 1.00	–	–	–	–
SD2, C3, I1	−1.8	0.4	0.4	0.5	0.27, 0.19	0.57, 0.63	0.91, 0.86	1.00, 0.98	1.00, 1.00	–	–	–	–
SD2, C4, I1	−1.8	0.4	0.4	0.7	0.22, 0.24	0.60, 0.62	0.92, 0.88	1.00, 0.97	1.00, 0.99	1.00, 1.00	–	–	–
SD3, C1, I1	−1.8	3.4	0.3	0	0.13, 0.01	0.39, 0.02	0.69, 0.07	0.89, 0.11	0.98, 0.18	1.00, 0.28	1.00, 0.34	1.00, 0.41	1.00, 0.54
SD3, C2, I1	−1.8	3.4	0.3	0.3	0.16, 0.01	0.43, 0.04	0.72, 0.07	0.91, 0.12	0.99, 0.20	1.00, 0.25	1.00, 0.36	1.00, 0.42	1.00, 0.48
SD3, C3, I1	−1.8	3.4	0.3	0.5	0.20, 0.01	0.48, 0.05	0.72, 0.09	0.90, 0.10	0.98, 0.18	1.00, 0.25	1.00, 0.35	1.00, 0.41	1.00, 0.55
SD3, C4, I1	−1.8	3.4	0.3	0.7	0.26, 0.02	0.51, 0.04	0.79, 0.06	0.90, 0.12	0.99, 0.19	1.00, 0.27	1.00, 0.33	1.00, 0.43	1.00, 0.53
SD3, C1, I2	−0.7	3.4	0.3	0	0.17, 0.02	0.52, 0.03	0.80, 0.07	0.96, 0.14	0.98, 0.19	1.00, 0.25	1.00, 0.34	1.00, 0.44	1.00, 0.54
SD3, C1, I3	0	3.4	0.3	0	0.17, 0.03	0.52, 0.08	0.79, 0.10	0.94, 0.14	1.00, 0.16	1.00, 0.26	1.00, 0.32	1.00, 0.41	1.00, 0.50
SD3, C1, I4	0.7	3.4	0.3	0	0.19, 0.04	0.47, 0.07	0.78, 0.10	0.92, 0.15	0.98, 0.22	1.00, 0.24	1.00, 0.31	1.00, 0.41	1.00, 0.51

Notes. SD=Standard Deviation; C=Correlation; I=Intercept

Table A2.

Multilevel Linear Regression Model

	Condition				Power for Specific Standardized Effect Sizes (Within-Person Effect, Between-Person Effect)
	Fixed Effect Intercept	Random Intercept SD	Random Slope SD Within-Person Predictor	Correlation between Random Intercept and Random Slope	0.1	0.2	0.3	0.4	0.5
SD1, C1	0	0.1	0.1	0	0.99, 0.98	1.00, 1.00	–	–	–
SD2, C1	0	0.4	0.4	0	0.43, 0.41	0.96, 0.92	1.00, 1.00	–	–
SD2, C2	0	0.4	0.4	0.3	0.42, 0.48	0.95, 0.95	1.00, 1.00	–	–
SD2, C3	0	0.4	0.4	0.5	0.40, 0.52	0.93, 0.95	1.00, 1.00	–	–
SD2, C4	0	0.4	0.4	0.7	0.42, 0.59	0.95, 0.99	1.00, 1.00	–	–
SD3’, C1	0	0.7	0.2	0	0.86, 0.19	1.00, 0.59	1.00, 0.87	1.00, 0.97	1.00, 1.00
SD3’, C2	0	0.7	0.2	0.3	0.86, 0.20	1.00, 0.62	1.00, 0.90	1.00, 0.98	1.00, 1.00
SD3’, C3	0	0.7	0.2	0.5	0.89, 0.20	1.00, 0.65	1.00, 0.91	1.00, 1.00	–
SD3’, C4	0	0.7	0.2	0.7	0.90, 0.23	1.00, 0.76	1.00, 0.99	1.00, 1.00	–

Notes. SD=Standard Deviation; C=Correlation; I=Intercept