Dyadic Patterns of Abstinence from Smoking in Dual-Smoking Couples Enrolled in a Pilot Randomized Clinical Trial

Abstract

Objective:

Despite the positive role of social relationships in behavior change, dyadic interventions for smoking have not been consistently successful. This owes, in part, to the fact that dual-smoking cohabiting couples share similar routines, environments, and relational dynamics that can undermine quitting efforts. We adopted an exploratory and descriptive approach to identify distinct profiles of smoking abstinence within dyads, how relationship dynamics relate to these patterns, and whether these patterns predict smoking status at follow-up.

Methods:

We utilized pre-existing data from a pilot RCT examining the effects of partner-involved financial incentives (PIFs) on dyadic abstinence. Both members of 95 dual-smoking couples (52% female, 73% White, M_age=42.39, SD=10.57) recruited between 2021-2022 provided baseline information about their relational and motivational dynamics, followed by 10 weeks of daily reports of smoking behavior, and biochemically verified quit status at follow-up.

Results:

Latent growth mixture models suggested four patterns of dyadic abstinence: Concordant abstainers, Discordant abstainers, Discordant non-abstainers, and Concordant non-abstainers. Treatment arm and motivational and relational dynamics predicted the probability of following specific dyadic patterns of behavior change, and dyadic patterns of change predicted individual and couple quitting status at follow-up.

Conclusions:

Our findings underscore the importance of considering that treatment will facilitate coordinated dyadic behavior change for some couples but not all and emphasize the need to better understand when and how mechanisms support long-term abstinence. More work is needed to investigate whether these patterns generalize across samples with more diverse sociodemographic and health characteristics.

Smoking is a highly concordant behavior among couples (Holahan et al., 2020; Tooley & Borrelli, 2017). In a recent meta-analysis of male-female partner correlations of 22 traits and behaviors, smoking status was among the strongest correlated traits between partners, behind only religiosity, political values, and educational attainment (Horwitz et al., 2023). Indeed, relationship partners commonly share habits, pursue similar goals, and intentionally and non-intentionally influence each other’s goal pursuits (Laurin et al., 2016; vanDellen et al., 2016). In dual-smoking couples, where both individuals smoke cigarettes, couple members tend to smoke more when together than apart, to become more likely to quit when one of the partners quits, and to desire partner’s involvement in quitting (Brazeau & Lewis, 2021; Holahan et al., 2020; Tooley & Borrelli, 2017; vanDellen et al., 2024). Dyad-level approaches may be especially important for populations such as dual-smoking couples where partners share the need for a specific health behavior change (Huelsnitz et al., 2022; vanDellen, 2019).

Dyadic models of self-regulation place health behavior change in the context of a system, wherein partners mutually influence each other’s behaviors and beliefs (Fitzsimons et al., 2015; Huelsnitz et al., 2022). In this interdependent and recursive network, partners allocate effort (e.g., support), set standards, monitor, and motivate each other’s progress toward personal, other-oriented, and shared goals (Fitzsimons et al., 2015; vanDellen, 2019). These relationship dynamics can be harnessed for successful quitting. For instance, support can increase efficacy, opportunities, and resources for behavior change, such that emotional and instrumental social support from one’s partner are associated with decreased probability of smoking and number of cigarettes smoked (Buitenhuis et al., 2021; Haskins et al., 2021; Lawhon et al., 2009; Lüscher et al., 2015). Relationship satisfaction can also facilitate cessation, potentially because it increases other means for successful goal pursuit, such as adoption of similar goals and mobilization of effort to support a partner’s goals (Foulstone et al., 2017; Hofmann et al., 2015; Pietromonaco & Collins, 2017; Toma et al., 2023).

Despite the potential positive role of relationships in behavior change, however, dyadic interventions for smoking have not generally been successful (Choi, 2022). There are many reasons why this may be the case. Most studies of smoking cessation in couples have focused on distal outcomes (i.e., abstinence at follow-up), potentially overlooking important dyad-level patterns of abstinence that influence the probability of long-term cessation. Additionally, relationship dynamics can promote maintenance of unhealthy behaviors. For example, health behaviors become increasingly concordant with increased contact, such as cohabitation and close relationships. Living with a smoker is associated with higher odds of future smoking (Holahan et al., 2020) and smoking couples share similar smoking routines, purchase packs of cigarettes for each other, bond over smoking together, are less likely to limit smoking at home, and experience dissatisfaction when one partner attempts to quit (Choi et al., 2020; Nyborg & Nevid, 1986; Shoham et al., 2007). From the perspective that partners’ goals do not function in isolation, but within an interdependent system, these findings suggest that quitting smoking may disrupt how couples live their lives together, making dyadic interventions both promising and challenging.

Present Study

The extant literature underscores the importance of understanding how dyadic behavior change unfolds over time. For instance, what patterns emerge in day-to-day abstinence among couples trying to quit smoking? In individual smokers, both intervention type (i.e., nicotine patches, lozenge, bupropion, or their combinations vs. placebo) and baseline psychosocial context (i.e., self-efficacy, racial and ethnic minority status, smoking history, sleep disturbance) predicted probability of early cessation rather than failure (McCarthy et al., 2015). However, dyadic patterns of smoking behavior change have never been examined. In this research, we utilized a pre-existing dataset reporting the results of a three-arm (two treatment conditions involving financial incentives; one control condition with no incentives) Phase IIa and b randomized efficacy and feasibility pilot RCT to examine these patterns.

Whereas past manuscripts using data from this RCT have combined the two treatment arms to focus on general effects of partner-involved financial incentives (PIFs; vanDellen et al., 2024; vanDellen et al., 2025), in this research, we take an exploratory approach to inspect patterns in daily abstinence data provided by both members of smoking couples, rather than the effects of the intervention. The treatment investigated in this trial showed preliminary efficacy, both in increasing quit attempts across 10 weeks (49.5%) and trial-wide abstinence rates (28%; vanDellen et al., 2024), which affords an opportunity to examine patterns of abstinence among initially smoking dyads who did make individual and joint quit attempts. Because of the presence of an intervention, we controlled all analyses for treatment arm. From this data, we aimed to understand patterns of dyadic abstinence that emerge during quit attempts, as well as whether patterns are related to treatment arm or dynamics in the couple.

In the present research, our focus was exploratory and descriptive. We examined dyadic patterns of abstinence during a 10-week period of weekly self-reported smoking behaviors between the delivery of intervention and a 3-month follow-up. We aimed to answer three research questions: (1) Is there evidence of distinct profiles of smoking abstinence within dyads? (2) Do relationship dynamics relate to patterns of smoking abstinence? (3) Do patterns of smoking abstinence relate to smoking status at follow-up? In light of exploratory aims, we did not specify directional hypotheses or run power analyses. Instead, we primarily focused on patterns of associations and descriptive statistics, given that standard errors and confidence intervals tend to be large in small samples such as this. Where possible, we report confidence intervals instead of p-values.

Method

Participants

Participants were both members of 95 (N=190, 52% female, M_age=42.39, SD=10.57) dual-smoking couples recruited between February 2021 and May 2022 and who provided online consent to be in this study. Most participants self-identified as White (73%), 12% as Black or African American, 11% as Multiethnic or Multiracial, and 1% as another race; 3% did not report a race or ethnicity. Participants included both opposite sex (n=83 couples) and same-sex couples (n=12 couples). Eligible participants had to smoke ≥5 cigarettes a day, be 18 years or older, have low risk for psychosis (<8 on the Psychosis Questionnaire [Ising et al., 2012]), and be romantically partnered and living with another eligible participant for 6+ months. Individuals with pregnancy or hospitalization within past 6 months were not eligible. Participants were recruited through advertisements on social media (Facebook, Instagram), ResearchMatch, and Craigslist across the United States. We refer to the first person to respond to advertising as the Target; the other as the Partner. Table 1 describes the baseline characteristics of all enrolled participants. Table S1 reports correlations between Target and Partner reports on all study variables (Target-Partner correlations ranged .20-.63). The study CONSORT diagram is available in the supplemental materials (see Figure S1). For additional information about recruiting and the parent trial, see vanDellen et al. (2024).

Table 1.

Descriptive Statistics for Baseline Measures

	n (%)	M	SD
Current smoking heaviness
5-10 cigarettes per day	54 (28.42)
10-20 cigarettes per day	90 (47.37)
More than 20 cigarettes per day	42 (22.11)
Did not report	4 (2.11)
Relationship length (in years; M, SD)		8.06	8.12
Relationship satisfaction (1-6)		3.25	0.46
Support for quitting – frequency (0-4)		0.99	0.83
Support for quitting – confidence (0-10)		6.23	2.76
Support for quitting – usefulness (0-10)		6.62	2.96
Negative influence (0-4)		0.91	0.88
Positive influence (0-4)		1.50	1.11
Target motivation for quitting a		−0.03	0.89
Target motivation for partner quitting a		−0.03	0.87

^aAverage of each item’s standardized response

Procedures

Enrolled participants completed a baseline video conference that included surveys and a biochemical test of smoking status, followed by 10 weeks of brief Timeline Follow-back surveys on past week’s smoking behavior, four weeks of optional online psychoeducation (with nicotine replacement therapy available upon request), and a follow-up video conference with surveys and another biochemical test. Biochemical test kits were mailed to enrolled participants before the baseline video conference. Replacement breath sensors were sent as needed for follow-up. Participants were randomly allocated to one of three conditions: No partner-involved financial incentives (no PIF; i.e., control; n=29 couples), partner-involved single-target financial incentives (i.e., PIF-ST; n=34 couples) in which only the first person to sign up for the study (i.e., Target) was eligible for financial incentives, and partner-involved dual-target financial incentives (PIF-DT; n=32 couples), in which both Target and their Partner were eligible. All participants earned up to $75 for completing the study, and treatment participants were eligible for a $100 incentive for completing the optional psychoeducation and a $100 incentive for being biochemically verified as quit at follow-up in PIF-ST and PIF-DT conditions. Regardless of condition, participants were not expected, prompted, or required to quit or set goals to quit to participate. All research activities were approved by The University of Georgia IRB (#2128).

Measures

Smoking Heaviness

Smoking heaviness was measured with one item from the Fagerstrom Test for Nicotine Dependence (Heatherton et al., 1991), which asked participants to report how many cigarettes they smoked per day (0=10 or less, 3=31 or more).

Relationship Satisfaction and Length

Relationship satisfaction was assessed with 31¹ of the 32-item Couple Satisfaction Index (Funk & Rogge, 2011), which assesses participants’ satisfaction with their relationships. The index utilizes seven parts, each with a different set of scales. Details about sections are provided in the supplement. Given different scaling in the original measure, all responses were z-scored and then averaged so higher scores indicate higher satisfaction (ω=.83²). Participants reported the length of their relationship in months and years (converted to months; r=.98).³

Motivation for Quitting

Motivation for quitting was assessed with four items each for (self) quitting and partner quitting. Items measured the importance of quitting (or partner quitting; 0=Not at all important, 10=Most important goal of my life), readiness to quit (or for partner to quit) within the next month (0=Not at all, 10=100% ready), confidence that they (or their partner) will quit within next month (0=Not at all, 10=100% confident), and motivation to quit smoking (or for partner to quit; 1=I don’t want [my partner] to stop smoking, 7=I really want [my partner] to stop smoking in the next month; Motivation to Stop Smoking Scale; Kotz et al., 2013). Given different scaling of the items, items were first standardized and then averaged. Higher scores indicate higher levels of motivation for quitting (ω=.88) or for partner quitting (ω=.85).

Partner Support for Quitting

Partner support for quitting was assessed with the 15-item Partner Support for Quitting Scale (Haskins et al., 2021), which gauges frequency of support (0=Never, 4=Very often; ω=.86), confidence that the couple can provide support (0=Not at all sure, 10=Extremely sure; ω=.91), and perceived usefulness of support (0=Not at all useful, 10=Extremely useful; ω=.95). Items were averaged within subscales. Higher scores indicate higher levels of each aspect of support.

Partner Influence

Positive and negative partner influence were gauged with the 20-item Partner Interaction Questionnaire (Cohen & Lichtenstein, 1990), which captures the frequency of negative (10 items; e.g., “Criticize your smoking”; ω=.94) and positive (10 items; e.g., “Compliment you on not smoking”; ω=.90) influence behaviors related to quitting (0=Never, 4=Very often).

Smoking Abstinence and Quitting

Daily abstinence from smoking was measured with weekly Timeline Followback surveys (Harris et al., 2009), in which participants reported smoking behavior (0=No, 1=Yes) over the past week (including day of survey). Follow-up quit status at the individual and couple levels was assessed through biochemically verified (via expired carbon monoxide) and self-reported smoking abstinence for 7 days or longer. Couples were categorized as either “both quit,” “one quit,” or “neither quit.”

Data Preparation

Missing responses on weekly surveys and at follow-up were treated as “did not abstain/quit” based on intent to treat analysis in RCTs (White et al., 2011). Before estimating structural models, we used confirmatory factors analyses (CFA) to evaluate the composite measures we created (described above) for target motivation to quit smoking, partner motivation to quit smoking, support for quitting, and partner influence. We used full-information maximum likelihood estimation to account for missing data. We evaluated fit with the comparative fit index (CFI), Tucker-Lewis fit index (TLI), and standardized root mean square residual (SRMR; McDonald & Ho, 2002) based on recommended criteria (Hu & Bentler, 1999)⁴.

Transparence and Openness

Our research questions and analytic approach were pre-registered and can be found along with data at https://osf.io/zhaub/ (Andrade & vanDellen, 2025).

Primary Analyses

To identify and evaluate the number of latent classes, we estimated unconditional growth mixture models over a 70-day period with 2-5 classes and intercepts and slopes for Targets and Partners. Given the low prevalence of abstinence during the first week of the study (6.32% overall), we also examined whether models with 63 days of categorical daily data (i.e., minus Week 1) improved classification quality and fit. The slopes of both trajectories were specified in equal increments so that the first day of the trajectory had a loading of 1 and the last of 0 (i.e., slopes reflect change from the start to the end of the 63- or 70-day period). We used multiple random starts to ensure replication of the best loglikelihood and that solutions did not converge at local maxima. A key concern in growth mixture models (GMM) is correct class enumeration (i.e., determining the final number of classes). There are no concrete guidelines on minimum sample sizes for appropriate enumeration, as simulation studies show that, beyond sample size, detecting the correct number of classes depends strongly on class separation (Nylund et al., 2007). To increase confidence in the use of GMM for the present analysis and the number of extracted classes, we estimated multiple fit statistics and indices for evaluating class enumeration. Simulations show that the Lo-Mendell-Rubin likelihood ratio test (LMR), and more so the bootstrap likelihood ratio test (BLRT), have sufficient power (>.80) to detect the correct k-class models with as few as 200 cases with binary outcomes (Nylund et al., 2007). Both tests are so powerful, however, that they can wrongly support models with incorrect number of classes (Liu & Hancock, 2014). We thus employed other methods for evaluating the number of extracted classes, including recommended BIC (Nylund et al., 2007; smaller values preferred) and entropy values (Wang et al., 2017; range 0-1, 1 indicates perfect separation). We favored models with class sizes that contained >5% of the sample (Weller et al., 2020). We used Mplus’ R3STEP to evaluate the effect of baseline covariates on the probability of class membership (RQ2) (Asparouhov & Muthén, 2014). We used the BCH method in Mplus (Asparouhov & Muthén, 2021) to examine separate models predicting the probability of Target, Partner, and Couple quitting as a function of class, both with conditional models (including baseline covariates) unconditional models (no baseline covariates) on quitting status (RQ3). In all hypothesis testing models, we used multiple imputation to account for missing data on baseline and follow-up variables.

Results

Confirmatory Factor Analyses of Measurement

We used confirmatory factor analyses (see Supplemental Table S2) to identify model fit when scales were treated as scales or subscales. Fit statistics suggested it was tenable to use composite indices for motivation for self and motivation for partner, subscale composites for partner support (usefulness, frequency, confidence) and partner interaction (positive, negative influence).

Is There Evidence of Distinct Dyadic Profiles of Smoking Abstinence Within Dyads?

As seen in Figure S2, Targets and Partners displayed similar prevalences of abstinence at the start of the monitoring period. Targets more commonly abstained than Partners, especially toward the end of the daily survey period.

Table S3 describes indices of model fit and classification quality for comparing models with 1-5 latent classes. Five-class models presented issues at several iterations that were not resolved by increasing the number of starting values. Entropy was adequate in all models and suggested high within-class homogeneity and between-class distinction. Although BLRT could not differentiate between neighboring class models, AIC, BIC, and aBIC indicated that a four-class solution was the best-fitting model, especially when isolating trajectories to Weeks 2-10. Additionally, the four-class solution isolating Weeks 2-10 yielded the lowest overall Pearson χ² and lowest percent of significant residuals, suggesting less misfit. Thus, we retained the four-class solution with Target and Partner growth trajectories from Weeks 2-10 for all subsequent analyses. Figure 1 displays the resulting fitted growth curves of Target’s and Partner’s daily probabilities of abstinence. Patterns of growth were used to label classes as Concordant abstainers, Discordant abstainers, Discordant non-abstainers, or Concordant non-abstainers.

Figure 1.

Fitted Growth Trajectories From Days 8 through 70 for Estimated Latent Classes

Table S4 reports, mean intercepts and slopes for Target and Partner by class. The Concordant abstainer class (12% of sample) was characterized by couples with a relatively higher probability of abstaining at the start of observation with both partners reaching a 99% probability of abstinence by end of Week 10. For both Targets and Partners, likelihood of abstinence significantly increased over time (positive slopes) with high likelihood of abstinence (vs. non-abstinence) by the last day of the tracking period (positive intercepts). The Discordant abstainer class (12% of sample) was characterized by couples with a low initial probability of abstaining and a slowly but significantly increasing probability of abstaining for both Targets and Partners (positive slopes). Both Targets and Partners became more likely to abstain over time, only Targets had a significantly higher likelihood of being abstinent than non-abstinent by the last day of tracking (significant intercept for Target, non-significant intercept for Partner).

Targets’ abstinence was more likely than Partners’ throughout. The Discordant non-abstainer class (16% of sample) was characterized by couples with members who started with differing probabilities of abstinence (19% Target; 46% Partner at start of Week 2). In this group, likelihood of abstaining from the start to the end of the tracking period did not change for Targets nor Partners in the Discordant non-abstainer class, and neither member of the dyad was no more likely to be abstinent than non-abstinent by the last day (non-significant intercepts and slopes). The Concordant non-abstainer class (61% of sample) was characterized by couples with a reliably low probability of abstaining across the period of observation. Targets and Partners in the Concordant non-abstainer class also were no more likely to be abstinent versus non-abstinent by the last day of the tracking period (non-significant intercepts), but Targets’ likelihood of abstaining increased from the start to the end of the tracking period (positive slope) whereas Partners’ did not.

Do Relationship and Smoking Dynamics Relate to Patterns of Smoking Abstinence?

Descriptive Differences by Class

Class membership did not differ as a function of relationship length, F(1,92)=1.60, p=.209. As seen in Figure 2, there were class differences in level of baseline variables across Target and Partner reports. Targets in Concordant and Discordant abstainer classes reported descriptively higher-than-average levels on all constructs except nicotine dependence. Targets in the Discordant non-abstainer class varied between above- and below-average levels on all constructs whereas Targets in the Concordant non-abstainer class were consistently below-average on all constructs, except nicotine dependence. See Table S5 for additional details.

Figure 2. Target’ and Partner’ Standardized Scores on Baseline Variables, by Class.

Note. Target’ and Partner’ standardized average responses on baseline constructs, by class. Averages were separately computed for Target and Partner and then standardized for each variable within each Target and Partner groups. Values above 0 reflect higher-than-average levels on that variable for that member of the dyad. Values below 0 reflect lower-than-average levels on that variable for that member of the dyad.

Patterns among Partners were relatively similar to those among Targets by class. Partners in the Concordant abstainer class reported higher-than-average levels on all constructs except relationship satisfaction. Partners in the Discordant abstainer class reported higher-than-average levels on motivation and support constructs, but below-average levels of positive and negative influence, relationship satisfaction, and nicotine dependence. Partners in the Discordant non-abstainer class reported higher-than-average levels on all constructs except relationship satisfaction and nicotine dependence. Partners in the Concordant non-abstainer class were below-average on all constructs, except nicotine dependence and relationship satisfaction.

Predicting Class Membership

We also formally examined whether baseline responses predicted the odds that a couple followed a specific pattern of abstinence over time (i.e., class membership) while controlling for condition. The full set of odds ratios and confidence intervals for pairwise comparisons between classes are displayed in Figure 3 and Supplemental Table S6. For each pairwise comparison, odds higher than 1 mean that higher levels on a variable increase couples’ odds of membership in the focal class relative to the comparison class. Conversely, odds below 1 mean that higher levels on a variable decrease the odds of membership in the focal class relative to the comparison class; seen differently, they increase the odds of membership in the comparison class.

Figure 3. Odds Ratios and 95% Confidence Intervals for Multinomial Logistic Regressions Predicting Probability of Latent Class Membership.

Note. Circles represent odds ratios; whiskers represent 95% confidence intervals. Each graph refers to the higher or lower odds (horizontal axis) of class membership (label before “vs.”) relative to the reference class (label after “vs.”) as Target- or Partner-reported levels on baseline variables (vertical axis) increase.

Effects of Condition on Class Status.

Couples in any incentive condition (PIF-ST, PIF-DT) had higher odds of being in the Concordant abstainer (OR_Target=4.99, 95% CI [1.63, 15.25], OR_Partner=3.02, 95% CI [1.17, 7.86]), Discordant abstainer (OR_Target=3.95, 95% CI [1.00, 15.55], OR_Partner=3.43, 95% CI [1.11, 10.61]), or Discordant non-abstainer (OR_Target=6.13, 95% CI [1.78, 21.13], OR_Partner=6.70, 95% CI [1.88, 23.84]) classes, relative to the Concordant non-abstainer class. Condition did not predict any other relative probability of class membership.

Patterns Related to Concordant Abstainer Class Status.

Couples in which Targets and Partners reported more frequent partner support had higher odds of being classified Concordant abstainer (OR_Target=6.90, 95% CI [1.22, 38.92], OR_Partner=10.37, 95% CI [1.16, 92.63]) instead of Concordant non-abstainers. More Target-reported relationship satisfaction (OR_Target= 19.03, 95% CI [2.58, 140.60]) and negative influence (OR_Target=4.76, 95% CI [1.53, 14.87]), as well as less positive influence (OR_Target=0.12, 95% CI [0.03, 0.46]), had higher odds of membership in the Concordant abstainer relative to Concordant non-abstainer class.

Patterns Related to Discordant Abstainer Class Status.

Longer relationship length (OR_Target=1.22, 95% CI [1.03, 1.43], OR_Partner=1.17, 95% CI [1.05, 1.29]) and less positive influence (OR_Target =0.12, 95% CI [0.02, 0.67], OR_Partner=0.04, 95% CI [0.00, 0.41]) increased odds of being in the Discordant abstainer relative to the Concordant non-abstainer class. Additionally, Target-reported partner support (confidence) increased odds of membership in the Discordant abstainer relative to Concordant non-abstainer class (OR_Target=4.81, 95% CI [1.17, 19.85]). Less Target-reported nicotine dependence and more negative influence predicted membership in the Discordant abstainer versus Concordant non-abstainer class (OR_Target=0.18, 95% CI [0.05, 0.68] and OR_Target=17.08, 95% CI [1.76, 166.08], respectively).

Patterns Related to Discordant Non-Abstainer Class Status.

More frequent support for quitting was associated with higher odds of being classified as Discordant non-abstainer versus Concordant non-abstainer (OR_Target=4.05, 95% CI [1.04, 15.84], OR_Partner=4.89, 95% CI [1.11, 21.58]). Less Target-reported partner support (confidence) had higher odds of being classified as Discordant non-abstainer versus Concordant (OR_Target=0.25, 95% CI [0.08, 0.77]) or Discordant abstainer (OR_Target=0.13, 95% CI [0.03, 0.53]). Lastly, higher relationship satisfaction had higher odds of being classified Discordant non-abstainer versus Concordant non-abstainer (OR_Target=7.21, 95% CI [1.28, 40.68], OR_Partner=0.11, 95% CI [0.02, 0.78]).

Do Patterns of Smoking Abstinence Relate to Smoking Status at Follow-up?

As Table 2 shows, probability of abstinence was high in the Concordant and Discordant abstainer classes for both Targets and Partners. We contrasted the relative probability of quitting as a function of pairwise class memberships (see Supplemental Tables S7 and S8). In unconditional models with no covariates, Target’s membership in the Concordant abstainer, Discordant abstainer, or Discordant non-abstainer classes were significantly associated with 98.38%, 99.26%, and 87.15% higher probability of quitting at follow-up compared to membership in the Concordant non-abstainer class (all p’s<.01). Likewise, Partner’s membership in the Concordant abstainer, Discordant abstainer, or Discordant non-abstainer classes were associated with 99.21%, 98.68%, and 94.95% higher probabilities of quitting at follow-up than membership in the Concordant non-abstainer class (all p’s<.01). Membership in the Concordant abstainer or Discordant abstainer classes for Targets and Concordant abstainer class for Partners was also significantly associated with a higher probability (89.96%, 95.22%, 87.03%) of quitting at follow-up relative to the Discordant non-abstainer class (all p’s<.05). With the exception of comparisons between Discordant non-abstainer and Concordant non-abstainer classes for Targets and Discordant non-abstainer and Discordant abstainer classes for Partners, these differences were robust to the addition of baseline covariates.

Table 2.

Probability of Target, Partner, and Couple Quitting at Follow-up by Class

		Probability	SE	p-value
Target	Concordant abstainer a	0.82	0.12	<.001
	Discordant abstainer a	0.91	0.09	<.001
	Discordant non-abstainer b	0.33	0.12	.006
	Concordant non-abstainer c	0.07	0.03	.038
Partner	Concordant abstainer a	0.82	0.12	<.001
	Discordant abstainer ab	0.73	0.13	<.001
	Discordant non-abstainer b	0.40	0.13	.002
	Concordant non-abstainer c	0.03	0.02	.150
Couple	Concordant abstainer a	0.82	0.12	<.001
	Discordant abstainer a	0.73	0.13	<.001
	Discordant non-abstainer b	0.27	0.12	.020
	Concordant non-abstainer c	0.02	0.02	.313

Note. Probabilities are based on most-likely class membership and do not account for classification error. Different superscripts indicate significant difference in unconditional probability of quitting (i.e., not accounting for covariates). Conditional probabilities and precise comparisons between groups are available in the Supplemental Materials (Tables S7 & S8).

Discussion

These results offer novel insights into how couples change smoking behavior over time, how the relationship context relates to dyadic patterns of day-to-day abstinence, and how these patterns relate to quitting status. In the present research, we observed four different patterns of dyadic abstinence. In two classes, individuals within dyad demonstrated strong behavioral concordance, either in becoming abstinent at similar times throughout the study or in persisting in regular smoking across the entire tracking period. Two classes showed more divergence in abstinence patterns. In one of these, both targets and partners became abstinent over time with partner abstinence rates lagging. In the other, targets’ probability of abstinence decreased over time while partner probability remained relatively low and did not change over time. These divergent patterns show that partner willingness to change may be a key factor driving target abstinence. Although in one class, partners’ abstinence was delayed, they have may expressed more support or motivation to quit or been preparing for a quit attempt. In the other class, partners may have provided negative influence or been stubborn about not quitting, contributing negatively to targets’ abstinence outcomes.

Predictors of class abstinence provide direction for future research to consider with larger samples and more precision through EMA designs. Generally, lower levels of nicotine dependence and a more positive relationship context predicted concordant dyadic abstinence over time. Target and Partner baseline levels of relationship and motivation factors were also descriptively more similar to each other among couples in the Concordant abstainer class, followed by those in the Discordant abstainer, then Discordant non-abstainer classes. This apparently shared level of enthusiasm, particularly in frequency of discussing abstinence strategies, may have created contexts that enabled abstinence (vs. non-abstinence) over time. Concordance was not inherently beneficial for behavior change, however. Specifically, some couples were concordant in ways that undermined their health (e.g., Concordant non-abstainers sharing low motivation) whereas others were concordant in ways that promoted health (e.g., Concordant abstainers sharing higher motivation to quit).

Interestingly, higher levels of negative influence (e.g., criticizing partner’s smoking) and lower levels of positive influence (e.g., [not] providing positive feedback when partner abstains) emerged as predictors of membership in one of the classes with higher probability of abstaining (i.e., Concordant abstainer, Discordant abstainer, and Discordant non-abstainers) relative to the Concordant non-abstainer class. Negative messages can backfire and decrease the effectiveness of quitting efforts (Scholz et al., 2020; vanDellen et al., 2019), however, they are also associated with concern for partner health and are more positively received between partners higher in relationship satisfaction (Di Maio et al., 2024; vanDellen et al., 2016). The effects of daily positive and negative messages may also differ when both partners (versus only one partner) want to change behavior, such that the joint effort and commitment may intensify beneficial effects and protect against negative effects of messages. Future work is needed to investigate whether effect of positive and negative messages differs when it comes to sustaining motivation during a quit attempt and whether this effect is influenced by other features of the relationship.

We also observed associations between being in an incentive condition and membership in an Abstainer class. This pattern could indicate that financial incentives can move the needle in the desired direction, but that concordance in partners’ motivation and relationship dynamics are also necessary for effective (and efficient) progress. This pattern is consistent with past evidence that couples with high levels of conflict responded more negatively to couple-framed efforts than couples with low levels of conflict (vanDellen et al., 2019). Inferences about concordance are descriptive-only in the present study, however, the fact that the same subset of Target-reported and Partner-reported baseline factors sometimes predicted probability of class membership suggests that future work is needed to test whether and how concordance relates to quitting.

Finally, dyadic patterns of abstinence throughout the study were associated with probability of quitting smoking at the couple and individual levels, lending empirical support for considering the temporal dynamics of dyads in the study of behavior change. Notably, couples in which one partner lagged behind the other and couples in which partners progressed at the same rate had a similar probability of quitting at follow-up, suggesting that dyadic trajectories and underlying processes can vary between couples yet yield similar outcomes. More interdependent couples may nevertheless be more prepared for dyadic interventions than less interdependent ones, as dyadic change may require that individuals make similarly timed efforts to quit smoking (Di Maio et al., 2024; Scholz et al., 2020). That is, couples in which both partners are motivated to quit and able to support another at the time of a quit effort may be best suited for dyadic interventions. Even in our research, lagged abstinence in Partners was temporally proximal to that of Targets, potentially due to congruence in Target and Partner motivation and support. Given that trial arm predicted class membership, and class membership predicted quitting at follow-up, the within-couple dynamics reflected in their trajectories (e.g., Target’s or Partner’s probability of smoking each day) may be a mechanism of change of dyadic incentive treatments. Further EMA studies are needed to test this possibility, adding to a growing interest in understanding how dyadic interventions influence behavior (Di Maio et al., 2024).

Limitations and Constraints on Generality

The trajectories observed in this study are similar to those in prior work with individual smokers (e.g., McCarthy et al., 2015) and consistent with evidence that health behavior tends to be concordant in couples (Brazeau & Lewis, 2021; Holahan et al., 2020; vanDellen et al., 2024). However, because this was a feasibility study, specific classes may be prone to variability in other trials, and observed patterns require further investigation and replication. For instance, the non-significant difference between Concordant and Discordant abstainers’ probability of quitting could stem from low power to detect one. The primary goal of this study was to identify and understand dyadic patterns of smoking abstinence. It is nevertheless possible that motivational and relationship dynamics shifted through the ten-week observation period or in tandem with dyadic patterns of abstinence, such that joint abstinence may have increased the positivity of the relationship context, increasing the probability of future abstinence (Fitzsimons et al., 2015; Laurin et al., 2016; Scholz et al., 2020). Future research could incorporate EMA designs to capture fluctuations in motivational and relationship dynamics along with daily abstinence.

Participants were not required to have or set quit goals, and thus this study may not reflect dyadic patterns that would emerge among initially motivated quitters. We also did not assess relationship, parenting status, or length of cohabitation, all of which could have directly or indirectly—through motivation, for example—influenced emergent classes and quitting status. Participants were, however, ineligible to enroll in the study if they smoked <5 cigarettes a day. Although this was intentionally done given our focus on daily smokers, it is possible that our findings do not translate to light smokers. Moreover, the dyadic nature of the intervention may have constrained relationship satisfaction levels. Enrollment favored non-Hispanic over Hispanic smokers, and underrecruited Native American populations. Additionally, given evidence that LGBTQ+ individuals have higher smoking rates than heteronormative individuals (Li et al., 2021), it was important to include LGBTQ+ couples in the study. Most couples were nevertheless heterosexual. Cultural differences surrounding smoking and relational processes in these undersampled populations may influence dyadic patterns of change. Finally, it is important to note these patterns were observed in the context of an effective cessation trial. Trial arm may have driven dyadic patterns of change and also have interacted with relationship context to predict patterns of abstinence and quitting status. For instance, that most couples classified as Discordant non-abstainers had been assigned to the dyadic-incentive condition could indicate that financial incentives alone are not sufficient to promote abstinence in partners with dissimilar motivation and relationship evaluations.

Conclusion

This study capitalized on daily abstinence data provided by both members of smoking couples to understand how dyadic behavior unfolds over time. Our findings show that couples can differ in their temporal patterns of abstinence and that the probability that a couple follows a given pattern is influenced by behavioral interventions as well as motivational and relationship dynamics. These dyadic patterns of abstinence over a nine-week period predicted individual and couple quitting status at follow-up. Our findings underscore the importance of considering that there may be multiple ways in which interventions change behavior in dyads, emphasizing the need to better understand when and how mechanisms support long-term abstinence.

Public Significance Statement.

This work advances a growing interest in understanding how dyadic interventions influence behavior change. This pilot study shows how motivational and relationship dynamics relate to patterns of behavior change over time. The most common patterns were similarly timed health behavior change within dyads and consistent lack of behavior change within dyads. We observed some flexibility in the extent to which couples’ motivational and relational dynamics predicted patterns of change and subsequent abstinence.