The Effects of Rumination, Distraction, and Gratitude on Positive and Negative Affect

Abstract

In a series of classic quasi-experiments, Lyubomirsky, Nolen-Hoeksema, and colleagues demonstrated the negative effects of rumination (vs. distraction) among people with depression. Across five studies, we attempted to replicate the former studies, as well as extend them by adding a third condition, gratitude, and the measurement of positive affect. We measured baseline depression severity and then randomly assigned people to a rumination, distraction, or gratitude condition. Pre- and post-manipulation, we measured depressed mood and positive and negative affect. We also explored whether manipulation-induced changes in affect related to construal of events, problem-solving, and thoughts about future behaviors (i.e., thought-action repertoires). As expected, both the distraction and gratitude conditions dampened negative affect (compared to the rumination condition), and the negative effects of the rumination condition were stronger among people relatively higher in baseline depression (compared to the distraction condition). Also as expected, the gratitude condition promoted positive affect when compared to the rumination and distraction conditions, an effect that was unmoderated by baseline depression. Furthermore, gratitude-induced changes in positive affect uniquely related to more positive construal of events, as well as higher positivity and lower negativity in thought-action repertoires. In sum, we found strong evidence that positive affect—above and beyond negative affect—facilitates healthy thought patterns, and we provide support for the idea that increasing positive affect should be a direct goal of treatments for depression, in addition to reducing negative affect.

In a series of classic quasi-experiments, Lyubomirsky, Nolen-Hoeksema, and colleagues demonstrated the negative effects of rumination among people with depression (Lyubomirsky & Nolen-Hoeksema, 1993, 1995; Lyubomirsky et al., 1998, 1999, 2003; Nolen-Hoeksema & Morrow, 1993). Specifically, they randomly assigned individuals with and without depression to either a self-reflective rumination condition or an externally-focused distraction condition and found that those with depression assigned to ruminate reported a host of negative outcomes compared to the other three groups (individuals with depression assigned to the distraction condition or individuals without depression assigned to the rumination or distraction conditions).¹ Rumination, compared to distraction, perpetuated depressive symptoms as the depressed ruminator prolonged or exacerbated their negative mood and reacted to the world in ways that fed back into their depression.

We revisited these classic studies in an attempt to replicate and extend their findings. Specifically, we were interested in whether a third condition designed to orient people to the positive in their lives (i.e., a gratitude condition) could potentially decrease negative affect among people with depression (compared to the rumination condition) and also increase positive affect, which might lead to additional benefits (Layous et al., 2014). Across five studies, we explored the effect of a rumination condition versus a distraction or gratitude condition on depressed mood, negative affect, positive affect, and various other outcomes. We also explored whether these condition effects were moderated by individual differences in baseline depression.

Rumination

According to Nolen-Hoeksema’s response styles theory, rumination (sometimes specified as depressive rumination) is a style of responding to one’s negative mood in which one repetitively and passively focuses on the meaning, causes, and consequences of the negative mood rather than acting to alleviate any of the circumstances surrounding it (Nolen-Hoeksema, 1991; Nolen-Hoeksema et al., 2008). Rumination is prospectively predictive of the onset of depressive symptoms (Lyubomirsky et al., 2015; Nolen-Hoeksema et al., 2008), and interacts with negative cognitive style to predict the duration of depressive episodes (Ciesla & Roberts, 2002; Robinson & Alloy, 2003). In laboratory studies, people with depression who ruminated (vs. distracted) maintained or exacerbated their depressed mood and also reported negative construal of hypothetical and real events in their lives (e.g., Lyubomirsky & Nolen-Hoeksema, 1995; Lyubomirsky et al., 1998, 1999), maladaptive patterns of problem solving (e.g., Lyubomirsky & Nolen-Hoeksema, 1995; Lyubomirsky et al., 1999), and impaired concentration (Lyubomirsky et al., 2003)—all negative thought patterns that could perpetuate their depression (Lyubomirsky & Nolen-Hoeksema, 1993). Importantly, Lyubomirsky, Nolen-Hoeksema, and colleagues’ rumination condition did not explicitly instruct people to focus on negative emotions or thoughts (e.g., “Think about the way you feel inside”), so, following their results, we only expected our rumination condition to be associated with higher depressed mood and negative affect among people with elevated baseline depression (see also Mor & Winquist, 2002).

Less research has focused on the effect of rumination on positive affect, but some evidence suggests that rumination will be related to lower positive affect. Indeed, in two experience-sampling studies, the use of rumination as an emotion regulation strategy at one time point was associated with decreases in positive affect at the next time point (Brans et al., 2013). Similarly, in a study of daily events, end-of-day reports of rumination were related to lower positive affect that day, and trait reports of rumination were also related to lower daily positive affect (Li et al., 2017, but see Chang, 2004 for a non-significant link). Furthermore, in an experiment following a sad mood induction, distraction and mindful self-focus conditions helped people regain positive affect, whereas a rumination condition did not, but was also not additionally harmful (Huffziger & Kuehner, 2009). In addition, induced rumination was associated with subsequent within-person reductions in positive mood, and this decrease was similar across levels of baseline depression (Huffziger et al., 2012). Thus, we expected that participants in our rumination condition would report lower positive affect than those in our gratitude condition (meant to actively foster positive affect), across levels of baseline depression.

Distraction

Nolen-Hoeksema and Morrow (1993) defined distraction as a response style to depressed mood in which people purposefully turn their attention away from their symptoms of depression on to pleasant or neutral thoughts and actions. By focusing people’s attention on something engaging and non-self-focused, distraction interrupts negative mood-congruent thinking, thereby dampening depressed mood (Erber & Tesser, 1992). As noted, Lyubomirsky, Nolen-Hoeksema, and colleagues’ studies demonstrated that a distraction condition improved depressed mood among individuals with depression compared to a rumination condition. Other studies have also demonstrated distraction’s beneficial effects on negative affect when compared to other conditions like mindful self-focus (distraction was as effective; Huffziger & Kuehner, 2009), happy memory recall (distraction was more effective; Joorman et al., 2007), or self-compassion (distraction was as effective; Odou & Brinker, 2015, but see Broderick, 2005 for evidence of mindfulness meditation being more effective than distraction).

Research on the effect of distraction on positive affect is more mixed. For example, in one of two experience-sampling studies, using distraction as an emotion-regulation strategy throughout the day was related to increases in positive affect, especially among people relatively higher on depression, but in the other study, there was no relationship (Brans et al., 2013). Following a sad mood induction, in one experiment, distraction improved positive affect as much as a mindful self-focus condition and more than a rumination condition (Huffziger & Kuehner, 2009); however, in another experiment, distraction decreased positive affect whereas self-compassion increased it (Odou & Brinker, 2015). In addition, following a positive (vs. neutral) mood induction, an engaging distraction task dampened positive mood compared to a less engaging task (Erber & Tesser, 1992). This evidence leaves open the possibility that distraction could either promote positive affect, dampen it, or have no effect. Given these mixed results, we expected that our distraction condition would not boost positive affect as well as a condition explicitly designed to do so—our gratitude condition. Therefore, we predicted that participants in our gratitude condition would report higher positive affect than those in our distraction condition across levels of baseline depression. Also, given these mixed results, and the minimal evidence regarding rumination and positive affect, we made no prediction regarding the potential differences in positive affect between the distraction and rumination conditions.

Importantly, even though distraction may dampen negative affect in the short-term, if it also dampens positive affect, it could undermine mental health or at least not help it long-term. Indeed, dampening positive affect is prospectively associated with higher depressive symptoms, even while controlling for ruminative responses to negative affect (Li et al., 2017; Raes et al., 2012). Similarly, although distraction effectively reduces negative affect in the short-term, a habitual practice of distraction—an avoidant coping strategy—relates to worse mental health outcomes in the long-term as people do not address the underlying issues for their negative affect. For example, one study found that avoidant coping strategies were related to higher depression ten years later, which was partially explained by more chronic and acute life stressors (Holahan et al., 2005; but see Huffziger et al., 2009). Thus, because distraction may not be a good strategy for long-term mental health, it is worth exploring other strategies, like gratitude, that may dampen negative affect compared to rumination, but also improve positive affect and have the potential for positive long-term outcomes.

Gratitude

Gratitude has been defined as one’s general tendency to appreciate the good things in their lives (i.e., as a trait, habit, moral virtue, or coping resource), as well as a transient emotion elicited from a particular situation or reflection in which one recognizes a positive personal outcome, not necessarily deserved or earned, that stems from another person or non-person entity (e.g., God or nature; Emmons & McCullough, 2003). Although Nolen-Hoeksema (1991) did not specifically consider gratitude under her response styles theory, she defined distraction as turning away from one’s depressive symptoms on to pleasant or neutral thoughts or actions. That said, the distraction condition in Lyubomirsky, Nolen-Hoeksema and colleagues’ studies focused on non-self-relevant neutral objects (e.g., “Think about and imagine a boat slowly crossing the Atlantic,” “Think about the shape of a large black umbrella”), not self-relevant positive ones. Instead of distracting participants with externally-focused, neutral topics, our gratitude condition redirects participants’ attention to thoughts that are self-focused and positive (e.g., “Think about the people in your life who value you,” “Think about someone who makes you feel special”). Gratitude has been implicated as an important coping resource (Fredrickson et al., 2003; Wood et al., 2007), and is associated with higher well-being and lower depression prospectively (e.g., Wood et al., 2008a; 2008b) and experimentally (Dickens, 2017; Cregg & Cheavans, 2021), so we wanted to compare the effects of a new gratitude condition to the existing rumination and distraction conditions among people with varying levels of depressive symptoms.

Given the anhedonic nature of depression, clinicians have noted some patients may have difficulty freely recalling things for which they grateful (Rashid & Seligman, 2018; see also Joormann et al., 2007 and Sin et al., 2011). Nevertheless, a strong body of research has found that gratitude interventions are beneficial among people with elevated depressive symptoms (e.g., Harbaugh & Vasey, 2014; Seligman et al., 2005; Sergeant & Mongrain, 2011), and even showed promise as a treatment among patients admitted to a psychiatric unit for suicidal thoughts or behaviors (e.g., Huffman et al., 2014). Indeed, a meta-analysis of 27 studies found that gratitude interventions successfully reduced symptoms of depression and anxiety, and the interventions were equally effective across levels of baseline depression severity (Cregg & Cheavens, 2021). Thus, we expected that the gratitude condition would encourage lower negative affect than the rumination condition, and we also explored whether this difference was moderated by baseline depression. In addition, as noted, we also expected the gratitude condition to promote higher positive affect than the rumination or distraction conditions.

We had two main reasons for employing a gratitude intervention instead of one of the comparison conditions that have already been shown to improve mood compared to a rumination or distraction condition (e.g., mindfulness meditation [Broderick, 2015], mindful self-focus [Huffziger & Kuehner, 2009], self-compassion [Odou & Brinker, 2015], self-affirmation [Koole et al., 1999]; positive reappraisal [Grisham et al., 2011; Rood et al., 2012]). First, none of these comparison conditions have the direct goal of upregulating positive affect, an important reason we think the gratitude condition will be a more effective emotion regulation strategy than the distraction condition. Instead, mindfulness and self-compassion encourage non-evaluative thoughts about one’s emotions, and self-affirmation is typically thought of as a way to protect one’s self-integrity in response to a threat. Similarly, positive reappraisal encourages people to think about a negative event in a different way, not to focus on positive events. Second, all of the aforementioned studies employed a negative induction prior to random assignment to condition (i.e., a sad mood induction or failure manipulation) and therefore it remains to be seen how these conditions would affect non-induced participant mood (like in the current studies). Nevertheless, these studies do suggest that some emotion regulation strategies may be more effective than distraction (e.g., Broderick, 2015; Odou & Brinker, 2015), and that positive affect may be an important tool in reducing rumination (e.g., Koole et al., 1999).

Lastly, people may wonder why we chose to employ a gratitude intervention rather than a different positive emotion induction (e.g., pride, relief, prosocial recall). Past research has found that gratitude inductions promote higher well-being than other positive affect inductions like recalling pride (Watkins et al., 2015), relief (Layous et al., 2017), or prosocial (Layous et al., 2017) experiences. Thus, we employed a gratitude intervention to have the best chance of boosting positive affect. In addition, Watkins et al. (2015) found that a gratitude induction also increased the accessibility of positive memories compared to pride and placebo conditions, indicating that gratitude may successfully induce positive cognitive biases that could counter the negative cognitive biases found in people with depression who ruminate.

Emotion Regulation Strategies Predict Thought Patterns via Negative and Positive Affect

In addition to our expectation that the rumination, distraction, and gratitude conditions will have differing effects on negative and positive affect, we also expected that these condition-induced differences in affect would predict different thought patterns. Lyubomirsky and Nolen-Hoeksema (1993) noted that people with depression who ruminated exacerbated their depressed mood and engaged in negative thought patterns that perpetuated their depression. In their series of experiments, they explored many negative thought patterns that we revisit in the current paper (i.e., negatively construal of hypothetical scenarios [Lyubomirsky & Nolen-Hoeksema, 1995]; higher self-reported frequency of negative events and lower self-reported frequency of positive events in one’s life [Lyubomirsky et al., 1998]; worse interpersonal problem solving [Lyubomirsky & Nolen-Hoeksema, 1995; Lyubomirsky et al., 1998]; and lower self-reported likelihood of engaging in potentially pleasant activities [Lyubomirsky & Nolen-Hoeksema, 1993]). Lyubomirsky, Nolen-Hoeksema, and colleagues measured depressed mood alongside these thinking outcomes but never tested whether rumination (vs. distraction) had an indirect effect on these outcomes via depressed mood (i.e., if depressed mood was the mechanism by which different emotion regulation strategies predicted different thinking patterns). As Teasdale (1983) noted, depressed mood and negative thinking form a vicious cycle that perpetuates and intensifies depression. We sought to capture part of this cycle—from emotion regulation strategy to affect to thought patterns. Specifically, we expected that the distraction (vs. rumination) and gratitude (vs. rumination) conditions would predict lower negative affect, which would then predict more positive thought patterns. Furthermore, instead of solely focusing on the explanatory value of negative affect in predicting thought patterns, we also wanted to explore the unique contribution of positive affect.

Research has found that mental health and mental illness constitute two separate unipolar dimensions (Keyes, 2005), as do their hallmarks, positive and negative emotions (Diener & Emmons, 1985). Despite the acknowledgement that positive and negative affect are independent and both important, researchers often focus on one or the other, failing to capture the complete picture of mental health (Keyes, 2007). Importantly, a trait tendency to experience positive affect predicts variance in depressive symptoms over and above a trait tendency to experience negative affect (Brown et al., 1998; Clark & Watson, 1991; Watson et al., 1988). Relatedly, a tendency to dampen one’s positive emotions predicts depression over and above a tendency to ruminate in response to one’s negative emotions, underlining the unique explanatory power of positive affect over and above negative affect (Raes et al., 2012). Positive affect may influence depression in a few ways—not only does positive affect mitigate the maladaptive effects of negative affect (Fredrickson et al., 2000; Riskind et al., 2013) and aid in coping with life’s stressors (Folkman, 2008; Folkman & Moskowitz, 2000; Fredrickson et al., 2003), but it also promotes positive thoughts and actions that can alleviate depression and promote flourishing (Fredrickson, 2001, 2013). In the current studies, we attempt to demonstrate that positive affect predicts differences in thought patterns even while controlling for negative affect.

The broaden-and-build theory of positive emotions explains how positive affect prompts positive thinking and behavior (Fredrickson, 2001, 2013). In regard to the broaden part of the hypothesis (most relevant to the current studies), positive affect signals to people that they are safe and do not need to attend to any immediate and specific threats, thus allowing them to broaden their thinking and attention, prompting creative solutions to problems. For example, people in a relatively positive mood are more efficient decision-makers (Isen & Means, 1983), often perform well on problem-solving tasks (e.g., Erez & Isen, 2002; Kavanagh, 1987), and are rated as more creative (e.g., Estrada et al., 1994). This broadened state also prompts the urge to act, play, and explore. In one study, participants induced into a positive (vs. neutral) mood came up with more overall responses—and more positive responses (i.e., higher urge to exercise and go outdoors)—to an open-ended task asking what they would like to do given their current emotional state (i.e., their “thought-action repertoires”), whereas those induced into a negative (vs. neutral) mood came up with fewer responses—and more negative ones (i.e., higher urge to be antisocial and lower urge to engage in schoolwork; Fredrickson & Branigan, 2005). In addition to the outcomes included by Lyubomirsky, Nolen-Hoeksema, and colleagues, we also included this last measure given its relevance to positive affect specifically (i.e., the Twenty Statements Test; Fredrickson & Branigan, 2005). Controlling for negative affect, we expected that gratitude (vs. rumination) and gratitude (vs. distraction) would predict higher positive affect, which would in turn relate to more positive thought patterns (e.g., more positive thought-action repertoires, better interpersonal problem-solving, less negative construal of hypothetical scenarios).

Although not a focus of the current studies, the “build” part of the broaden-and-build hypothesis posits that, while in this positive state of openness and creativity, people may take numerous actions that could build lasting personal resources and foster long-term well-being (e.g., reaching out to a friend, starting a gym routine, or applying for a new job; Fredrickson et al., 2008). Thus, positive affect prompts positive thoughts and actions that better people’s lives, feeding back into well-being in an upward spiral (Fredrickson, 2001, 2013). Indeed, cross-sectionally, prospectively, and experimentally, positive affect predicts positive outcomes in career, physical health, and relationships (Lyubomirsky et al., 2005; Pressman & Cohen, 2005) which further reinforce mental health. Thus, it is critical to understand how positive affect, in addition to negative affect, might be induced to promote positive outcomes in daily living.

In the current studies, by including both negative and positive affect as mechanisms, we are not only capturing part of the vicious cycle described by Teasdale (1983)—from emotion regulation strategy to negative affect to negative thought patterns, but also part of the virtuous cycle described by Fredrickson (2001, 2013)—from emotion regulation strategy to positive affect to positive thought patterns. If the current studies demonstrate that (1) distraction and gratitude both downregulate negative affect (compared to rumination), (2) gratitude upregulates positive affect (compared to rumination and distraction), and (3) positive affect uniquely predicts more positive thought patterns, we will have further evidence for the need for therapeutic practices to focus on addressing both positive and negative affect, rather than assuming a reduction in negative affect will also address anhedonia and its effects.

The Current Studies

Across five studies, we sought to replicate and extend prior findings by Lyubomirsky, Nolen-Hoeksema, and colleagues (Lyubomirsky & Nolen-Hoeksema, 1993, 1995; Lyubomirsky et al., 1998, 1999, 2003; Nolen-Hoeksema & Morrow, 1993) on the maladaptive effects of rumination (vs. distraction) among people with relatively higher baseline depression. In addition to the neutral distraction comparison group, we added a gratitude condition meant to not only downregulate negative emotions, but also upregulate positive ones when compared to the rumination condition (and upregulate positive affect when compared to the distraction condition). Our goal was to explore whether gratitude could be a preferable emotion regulation strategy to distraction given its ability to increase positive affect, as well as decrease negative affect. Correspondingly, we predicted that increases in positive affect and decreases in negative affect would uniquely predict more positive thought patterns on our outcome measures (e.g., more effective problem solving, more positive thought-action repertoires). That is, we expected that increases in positive affect would be additionally helpful in that they would relate to healthier thought patterns above and beyond decreases in negative affect. Furthermore, although we lacked sufficient empirical evidence to form a priori hypotheses about other potential between-condition comparisons and related interactions, we examined these associations on an exploratory basis to gather additional information on our variables of interest. We outline our a priori hypotheses and exploratory analyses below.

Distraction vs. Rumination

Hypothesis 1a.

Among people relatively higher in baseline depression, those in the distraction condition will report lower depressed mood than those in the rumination condition (i.e., we will find a distraction [vs. rumination] by baseline depression interaction on depressed mood).

Hypothesis 1b.

Among people relatively higher in baseline depression, those in the distraction condition will report lower negative affect than those in the rumination condition (i.e., we will find a distraction [vs. rumination] by baseline depression interaction on negative affect).

Distraction vs. Rumination Exploratory Analyses.

We will explore whether participants in the distraction condition report different levels of positive affect than those in the rumination condition, and also whether these differences vary by baseline depression (i.e., is there a distraction [vs. rumination] by baseline depression interaction on positive affect?).

Gratitude vs. Rumination

Hypothesis 2a.

Participants in the gratitude condition will report lower depressed mood than those in the rumination condition.

Hypothesis 2b.

Participants in the gratitude condition will report lower negative affect than those in the rumination condition.

Hypothesis 2c.

Participants in the gratitude condition will report higher positive affect than those in the rumination condition.

Gratitude vs. Rumination Exploratory Analyses.

We will explore whether any of the above differences between the gratitude and rumination conditions vary by level of baseline depression (i.e., is there a gratitude [vs. rumination] by baseline depression interaction on depressed mood, negative affect, or positive affect?).

Gratitude vs. Distraction

Hypothesis 3.

Participants in the gratitude condition will report higher positive affect than those in the distraction condition.

Gratitude vs. Distraction Exploratory Analyses.

We will explore whether participants in the gratitude condition report different levels of depressed mood or negative affect than those in the distraction condition. In addition, we will explore whether these differences on depressed mood, negative affect, or positive affect vary by level of baseline depression (i.e., is there a gratitude [vs. distraction] by baseline depression interaction on depressed mood, negative affect, or positive affect?).

Indirect Effects Analyses

Hypothesis 4a.

The distraction and gratitude conditions will predict lower negative affect than the rumination condition, which will in turn predict more positive thought patterns (i.e., less negative and more positive thought-action repertoires, less negative construal of hypothetical scenarios, lower frequency of recalled negative events, higher frequency of recalled positive events, more effective interpersonal problem solving, more positive evaluation of pleasant activities).

Hypothesis 4b.

The gratitude condition will predict higher positive affect than the rumination and distraction conditions, which will in turn relate to more positive thought patterns.

Method

The Institutional Review Board at California State University, East Bay (CSUEB-IRB-2015–293-F; CSUEB-IRB-2016–255-F) approved our study protocols. The five presented studies all included depressed mood and positive and negative affect as outcomes and only varied with respect to the non-affect outcomes included (i.e., the specific thought pattern outcome). Thus, we include one method and results section to reduce redundancy and highlight similarity in results across studies. Studies are numbered in order of data collection and all data exclusions are disclosed in Supplemental Material.

Transparency and Openness

We reported how we determined our sample size, all data exclusions, all manipulations, and all measures in the study. All data, analysis code, and research materials are available at https://osf.io/mnex6/, and we followed JARS (Applebaum et al., 2018). Most data analyses were conducted using SPSS version 24, as well as the PROCESS macro version 3.3 (Hayes, 2018). Meta-analyses on affect outcomes were conducted using metacor within the meta package (version 5.2–0; Schwarzer, 2022; Schwarzer et al., 2015) in R version 4.1.3 (2022–03-10). The design and analyses from these studies were not pre-registered. In addition to the main manuscript, we also submitted two Supplemental Material documents, one with follow-up analyses referred to in the main manuscript and one with the manipulations and measures from all five studies.

Power Analyses

All of our primary analyses are regression-based. According to Cohen (1988), standardized betas of .14, .39, and .59 correspond to small, medium, and large effect sizes, respectively. We thought that the comparison between the gratitude and distraction condition on positive affect may be the smallest effect and similar to other experiments in which positive psychological interventions (e.g., gratitude interventions) show small (Bolier et al., 2013) to medium (Sin & Lyubomirsky, 2009) average effect sizes when compared to control conditions. Thus, we powered our study to detect a standardized beta of .26, which is approximately halfway between the values for small and medium effects (Fritz & MacKinnon, 2007).

We conducted a power analysis in G*Power 3.1 (F-test—Linear multiple regression: Fixed model, R² increase option; Faul et al., 2009). We set Cohen’s f² to .0725 (f² = R²/[1-R²]; .26²/[1-.26²]) to estimate the necessary sample size for achieving 90% power to detect a small-to-medium effect of our five predictors (distraction vs. rumination; gratitude vs. rumination; baseline depression, two interaction terms) at an appropriate level of significance, controlling for baseline levels of the outcome when necessary. At α = .05, we needed a sample of at least 233 participants, which we exceeded in all five studies. Importantly, in all five studies, we ran our regression analyses on multiple outcomes, but did not correct for these multiple comparisons. Given this is the first test of a new gratitude condition, and among the first tests of the rumination and distraction conditions on positive affect, we emphasized discovery. When prioritizing discovery, being too conservative is more concerning than not being conservative enough (Rothman, 1990; see also Rubin, 2021 for an argument against automatic adjustment for multiple comparisons). That said, most of our outcomes appear in at least two studies, and the affect outcomes appear in all five studies, allowing us a chance to replicate our findings and perform a meta-analysis across these effects.

Participants

Student Samples

Participants from Studies 1 (N = 298), 2 (N = 269), and 4 (N = 278) were introductory psychology students at California State University, East Bay who received course credit in exchange for participation during Winter quarter 2016, Spring quarter 2016, and Winter quarter 2017, respectively. Demographics from Study 1 are reported here and are similar to Studies 2 and 4. The sample was comprised of mostly women (67.1% women; 32.9% men). Ages ranged from 18 to 46 years (M = 19.89, SD = 3.05). Students identified their races and ethnicities as follows: 35.9% Hispanic/Latinx, 21.8% Asian, 14.8% Black/African American, 11.1% White, 11.1% “more than one,” 3.0% Hawaiian/Pacific Islander, and 2.3% “other.”

Online Samples

For Study 3, participants were recruited via the “Panels” feature on Qualtrics during October and November of 2016 and earned $5.00 for completion of each of two waves of the study (about 30 minutes each) for a potential total of $10.00. Because we had a final goal of having 300 participants complete a second time point, Qualtrics set a goal to recruit at least 600 participants for the first time point, assuming large attrition. A total of 642 participants provided complete data and valid attention checks at Time 1 (see data exclusion and attrition information in Supplemental Material). These participants ranged in age from 18 to 78 years old (M = 48.27, SD = 12.91). The majority of participants identified as White (86.1%), followed by 4.7% Black/African American, 3.9% Asian, 3.3% Hispanic/Latinx, 1.1% “more than one,” 0.5% American Indian/Alaskan Native, 0.3% “other,” and 0.2% Hawaiian/Pacific Islander. Because of Qualtrics sampling procedures, we had a relatively even breakdown of men (55.5%) and women (44.1%), and 0.5% of the sample identified as nonbinary. About half of the original participants also completed our second time point, but, due to the largely nonsignificant findings out to Time 2, we reserve those findings for Supplemental Material.

For Study 5, mTurk workers earned $5.00 in exchange for 30 minutes of their participation during September 2017 (N = 463; 58.2% identified as men, 41.3% as women, and 0.4% as nonbinary). Participants’ ages ranged from 19 to 73 years (M = 34.83, SD = 9.80) and they identified their races and ethnicities as follows: 74.0% White, 8.2% Black/African American, 7.8% Asian, 6.9% Hispanic/Latinx, 1.9% “more than one,” 0.4% “other,” 0.4% American Indian/Alaskan Native, and 0.2% Hawaiian/Pacific Islander.

Measures

We provide complete materials for Study 1 through 5 in Supplemental Material.

Baseline Depression

In all five studies, we assessed baseline depression severity with the Beck Depression Inventory-II (BDI-II; Beck et al., 1996a, 1996b), which asks participants to consider 21 groups of statements—each statement representing different degrees of severity on a particular symptom of depression (e.g., sadness, self-dislike, sleeping patterns, appetite, loss of interest)—and select which statement best represents how they have been feeling over the past two weeks. We excluded the suicide question for ethical reasons, so included 20 items. Most items included four statements with a score of 0 indicating least severity and a score of 3 indicating most severity (e.g., sadness: 0 = I do not feel sad; 1 = I feel sad much of the time; 2 = I am sad all the time; 3 = I am so sad or unhappy that I can’t stand it). We summed across the 20 statements to calculate a total depression score for each participant (Study 1: Range: 0–40, M = 12.84, SD = 8.60; Study 2: Range: 0–40, M = 13.11, SD = 8.43; Study 3: Range: 0–56, M = 3.67, SD = 6.00 [n = 272 reported their score as zero which was 42.37% of the sample]; Study 4: Range: 0–46, M = 13.21, SD = 7.76; Study 5: Range: 0–57, M = 9.72, SD = 11.14; αs across studies > .86). Typically, scores of 0 to 13 indicate minimal depression, scores of 14 to 19 indicate mild depression, scores of 20 to 28 indicate moderate depression, and scores of 29 and more indicate severe depression (Beck et al., 1996a, 1996b).

Affect

In all five studies, before and after the manipulation, participants were asked to report the degree to which they felt 27 items “RIGHT NOW” (e.g., sleepy, agreeable, sad, happy) rated on a 7-point scale (1 = not at all to 7 = extremely). To replicate Lyubomirsky, Nolen-Hoeskema, and colleagues’ work, we formed a depressed mood composite by combining the items sad and depressed/blue (α > .83 at pre- and post-test across all studies). To form a negative affect composite, we averaged the items unhappy, frustrated, angry/hostile, worried/anxious, and guilty (the first four listed items are from the Affect-Adjective Scale; Diener & Emmons, 1985; α > .77 at pre- and post-test across all studies). We also combined the items happy, pleased, joyful, and enjoyment/fun to create a positive affect composite (Diener & Emmons, 1985; α > .86 at pre- and post-test across all studies). In Supplemental Material, we also include analyses conducted on a gratitude composite formed by combining the items thankful, grateful, and appreciative (Emmons & McCullough, 2003), and on the single-item indebtedness. In addition, we included 12 filler items to throw off the true purpose of our study (e.g., sleepy, curious, careful).

Twenty Statements Test

In Studies 1 and 3, participants described, in a word or two, the strongest emotion they felt while engaging in the experimental manipulation (Fredrickson & Branigan, 2005; modified from Kuhn & McPartland, 1954). Next, participants concentrated on the emotion and felt it as vividly and deeply as possible. Given this feeling, participants listed all of the things they would like to do “right now” (i.e., their thought-action repertoires) by responding to open-ended sentence stems that started with, “I would like to _____.” Participants could list up to 20 responses, but they could stop whenever they ran out of ideas. Because past research has found that people in a positive state provided more responses than those in a neutral or negative state (Fredrickson & Branigan, 2005), we recorded the number of responses participants completed out of 20 (Study 1: M = 7.97, SD = 5.51; Study 3: M = 4.85, SD = 5.24).

Next, in each study, two independent coders, blinded to condition, read participants’ complete set of responses and evaluated the degree to which the thought-action repertoires were positive (1 = not at all positive to 7 = very positive; Study 1: M = 2.64, SD = 1.44; Study 3: M = 4.12, SD = 1.71) and negative (1 = not at all negative to 7 = very negative; Study 1: M = 1.42, SD = 0.93; Study 3: M = 1.46, SD = 1.06) in two separate variables. Absolute agreement was moderate to high between the raters on both variables in both studies and we used the average of their ratings for all analyses (Study 1: positivity: ICC[2, 2] = 0.75; negativity: ICC[2, 2] = 0.78; Study 3: positivity: ICC[2, 2] = .56; negativity: ICC [2, 2] = .76; Fleiss, 1986; Shrout & Fleiss, 1979). In Study 1, only one person provided zero responses, so all participants were coded for negativity and positivity. In Study 3, 41 people gave zero responses and thus they were excluded from analyses regarding the negativity and positivity ratings.

In addition, the independent raters categorized each of the 20 responses into one of 16 (Study 1) or 17 (Study 3) mutually exclusive categories representing what the participant said they wanted to do at that moment (adapted from Fredrickson & Branigan, 2005; e.g., sleep/rest, be social, play). Analyses on these specified categories by condition and baseline level of depression are included in Supplemental Material.

Cognitive Bias Questionnaire

In Studies 1 and 5 participants completed The Cognitive Bias Questionnaire which is a measure of depressive thinking that presents six scenarios, followed by a series of questions designed to elicit participants’ thoughts and feelings in reaction to the scenarios (Hammen & Krantz, 1975; Krantz & Hammen, 1979; see its use in Lyubomirsky & Nolen-Hoeksema, 1995). Specifically, participants pictured themselves in the place of the person in the scenario and imagined what that person thought and felt. Each question had four possible answers in a multiple-choice format, one of which was depressed and distorted (i.e., negatively biased), one that was nondepressed but distorted, one that was depressed but not distorted, and one that was nondepressed and nondistorted. For example, in one scenario, a person in a community organization ran for the presidency but lost. Participants were asked, “When you first heard you’d lost, you immediately: (a) Feel bad and imagine I’ve lost by a landslide (depressed/distorted), (b) Shrug it off as unimportant (nondepressed/distorted), (c) Feel sad and wonder what the total counts were (depressed/nondistorted), or (d) Shrug it off, feeling I’ve tried as hard as I could (nondepressed/nondistorted).” There were 23 questions total. The number of depressed/distorted (Study 1: M = 2.49, SD = 2.57; Study 5: M = 2.43, SD = 2.93) and nondepressed/nondistorted (Study 1: M = 11.97, SD = 3.89; Study 5: M = 11.71, SD = 4.69) responses were summed across the questions to get participants’ total of negatively biased (depressed and distorted) and neutral (nondepressed/nondistorted) responses. Responses that were depressed/nondistorted or nondepressed/distorted did not contribute to either sum.

Event Frequency Task

In Studies 2 and 3, participants were presented with a list of 20 events and asked to rate how frequently these events happened to them in their life (1 = never, 2 = rarely, 3 = occasionally, 4 = frequently, 5 = always; Lyubomirsky et al., 1998). Ten of these events were rated by nondepressed coders as negative (e.g., “you receive unfair treatment”) and ten were rated as positive (e.g., “a close family member expresses love to you”; Lyubomirsky et al., 1998). We averaged the ten negative events (Study 2: M = 2.88, SD = 0.55, α = .74; Study 3: M = 2.57, SD = 0.71) and ten positive events (Study 2: M = 3.38, SD = 0.56, α = .74; Study 3: M = 3.13, SD 0.77) into two separate composites.

Means-Ends Problem Solving Task

In Study 2, participants completed The Means-Ends Problem Solving Task, which is a measure of interpersonal problem solving in which participants see the beginning and ending of four hypothetical interpersonal problems and fill in the middle of the story by describing what they did to get from problem to resolution (Lyubomirsky & Nolen-Hoeksema, 1995 adapted from Platt & Spivack, 1975 to better fit nonclinical undergraduate samples; see also its use in Lyubomirsky et al., 1998). For example, in one scenario, participants imagine one of their close friends is avoiding them. The situation ends when the friend likes them again.

Lyubomirsky and Nolen-Hoeksema (1995) presented each of the four situations to eight independent judges and asked them to list “model” solutions. There was a high degree of consensus among judges about model solutions. For example, for the friend scenario mentioned above, model solutions included going to see the friend in person, approaching the issue in a tactful way, and saying something to reaffirm the friendship. Non-model solutions included avoiding the friend, acting mean or insensitive toward the friend, and blaming or criticizing the friend when discussing the issue.

Four independent coders who were blind to condition rated all participants on two or more of the stories. Each story was rated on the overall effectiveness of the response (1 = not at all effective; 4 = moderately effective; 7 = extremely effective) and the ratings of the two coders who agreed the best were averaged (Story 1: ICC[2,2] = .61; M = 4.53, SD = 1.51; Story 2: ICC[2,2] = .67; M = 4.08, SD = 1.60). Adequate agreement was not reached for the overall effectiveness of Story 3 (ICC[2,2] = .19) or Story 4 (ICC[2,2] = .34) so those stories are not analyzed here (Fleiss, 1986). Coders also counted the number of model solutions in each story (ICC[2,2] > .54), as well as the number of total solutions (ICC[2,2] > .58), and the percentage of effective solutions out of the total was calculated for each story and then averaged across the four stories (M = 0.74, SD = 0.16).

Judgment of Pleasant Activities

In Study 4, following Lyubomirsky and Nolen-Hoeksema (1993), we asked participants to consider 24 activities and to answer the following two questions: “How enjoyable would you find this activity?” (5-point scale: 1 = not at all enjoyable, 2 = slightly enjoyable, 3 = somewhat enjoyable, 4 = moderately enjoyable, 5 = extremely enjoyable) and “How likely would you be to engage in this activity if you had the opportunity?” (5-point scale: 1 = not at all likely, 2 = slightly likely, 3 = somewhat likely, 4 = moderately likely, 5 = extremely likely). These activities had been previously judged by students in the first author’s experimental psychology class to be pleasant and distracting for other undergraduate students. Examples include, “Spending time with friends,” “Binge watching a good TV show,” and “Going out to eat.” The 24 “enjoy” and “likely” questions were averaged into separate composites (enjoy: M = 4.03, SD = 0.40, α = .77; likely: M = 3.89, SD = 0.47, α = .81).

Procedure

Our student samples (Studies 1, 2, and 4) from the psychology participant pool signed up for a time and location to engage in a 50-minute study titled “Cognition Studies.” In the study session, research assistants from the first author’s research methods course provided participants with consent information and told participants that they would be engaging in a series of short, independent studies from different researchers to make the most of their time. Research assistants said that all of the studies involved “processes of imagination, dreaming, cognition, and personality.” After the introduction to the experiment, participants provided consent and then started the study while seated at a computer queued to the study website. Some participants completed the study in a room with eight separate experimental rooms, and others completed the study in a room with four computers all in the same room. Participants from our online samples (Studies 3 and 5) were recruited from their respective online platforms, and did not appear in person, but otherwise followed the same study procedure as our student samples.

The first part of the experiment was labeled “Part A” and it entailed completion of four filler scales, the Beck Depression Inventory (BDI), the baseline affect measure, and two other measures not analyzed in the current paper (i.e., Satisfaction with Life Scale, Diener et al., 1985; Subjective Happiness Scale, Lyubomirsky & Lepper, 1999). Next, participants proceeded to “Part B,” which was purported to be a new study. At this point, Qualtrics randomly assigned participants to one of three conditions: rumination, distraction, or gratitude using the “evenly present elements” option (sample sizes by condition and study are included in Table 1). Complete instructions are available in Supplemental Material.

Table 1.

Means and Standard Deviations of Affect by Condition at Pre- and Post-Test

Condition	Depressed Mood				Negative Affect				Positive Affect
	Pre-Test		Post-Test		Pre-Test		Post-Test		Pre-Test		Post-Test
	n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)
Study 1
Rumination	99	1.95 (1.26)	99	1.91 (1.19)	99	2.08 (1.14)	99	2.06 (1.16)	99	3.69 (1.63)	99	3.62 (1.67)
Gratitude	99	2.12 (1.42)	99	1.83 (1.21)	99	2.18 (1.11)	99	1.89 (1.03)	99	3.49 (1.35)	99	3.98 (1.45)
Distraction	100	2.17 (1.49)	100	1.88 (1.32)	100	2.35 (1.22)	100	1.94 (1.11)	100	3.99 (1.54)	100	3.65 (1.56)
Study 2
Rumination	91	2.01 (1.22)	91	2.16 (1.38)	91	2.13 (0.95)	91	2.16 (1.12)	91	3.71 (1.47)	91	3.59 (1.48)
Gratitude	89	1.96 (1.19)	89	1.92 (1.23)	89	2.04 (0.95)	89	1.79 (0.86)	89	3.69 (1.46)	88	4.12 (1.65)
Distraction	89	2.21 (1.49)	89	1.93 (1.34)	89	2.31 (1.17)	89	2.06 (1.15)	89	3.60 (1.45)	89	3.46 (1.54)
Study 3
Rumination	203	2.26 (1.64)	203	2.24 (1.67)	203	2.34 (1.52)	203	2.18 (1.43)	203	3.64 (1.69)	202	3.61 (1.75)
Gratitude	229	2.24 (1.60)	229	2.11 (1.48)	229	2.20 (1.34)	229	1.97 (1.27)	229	3.50 (1.70)	229	3.68 (1.78)
Distraction	210	2.16 (1.79)	210	1.92 (1.58)	210	2.24 (1.47)	210	2.00 (1.45)	210	3.75 (1.77)	210	3.81 (1.84)
Study 4
Rumination	92	1.97 (1.40)	92	2.08 (1.44)	92	2.05 (1.09)	92	1.97 (1.04)	92	3.39 (1.20)	92	3.20 (1.42)
Gratitude	93	2.22 (1.49)	93	1.94 (1.16)	93	2.32 (1.16)	93	1.76 (0.83)	93	3.65 (1.27)	93	3.97 (1.50)
Distraction	93	1.94 (1.18)	93	1.61 (1.03)	93	2.00 (1.07)	93	1.71 (0.86)	93	3.38 (1.30)	93	3.26 (1.50)
Study 5
Rumination	156	1.80 (1.33)	156	1.96 (1.44)	156	1.80 (1.11)	156	1.93 (1.18)	156	3.16 (1.61)	156	3.20 (1.68)
Gratitude	151	1.95 (1.31)	151	1.92 (1.39)	151	1.93 (1.15)	151	1.77 (1.07)	151	3.31 (1.60)	151	3.91 (1.74)
Distraction	156	1.72 (1.15)	156	1.78 (1.32)	156	1.70 (0.98)	156	1.63 (0.93)	156	3.49 (1.71)	156	3.46 (1.74)

In all three conditions, participants focused on a series of 45 statements and could not advance to the next screen until eight minutes had passed. In the rumination condition, the statements were self-focused, emotion-focused, and symptom-focused, but participants were not explicitly told to focus on negative emotions or thoughts. For example: “Think about: your character and who you strive to be,” “Think about: the possible consequences of your current mental state,” or “Think about: what it would be like if your present feelings lasted” (Nolen-Hoeksema & Morrow, 1993). In the distraction condition, participants focused on non-self-relevant, neutral content such as, “Think about: and imagine a boat slowly crossing the Atlantic,” “Think about: the shape of a large black umbrella,” or “Think about: the movement of an electric fan on a warm day” (Nolen-Hoeksema & Morrow, 1993). Based on the format of the rumination and distraction conditions from Nolen-Hoeksema and Morrow (1993), we created the gratitude condition by using self-relevant statements focused on the positive aspects of one’s life. For example, “Think about: the people in your life who value you,” “Think about: the way you feel when you’re with your closest friends or family,” or “Think about: a beautiful place in nature.” After participants completed this task, they again reported their current affect and then completed the Twenty Statements Test (Studies 1 and 3). Finally, participants advanced to “Part C,” in which they completed any additional outcome measures (Cognitive Bias Questionnaire [Studies 1 and 5]; Event Frequency Task [Studies 2 and 3], Means-Ends Problem Solving Task [Study 2], Judgment of Pleasant Activities [Study 4]), demographics, and then read a debriefing statement.

Our procedure differed from Lyubomirsky, Nolen-Hoeksema and colleagues’ studies in a few ways. First, in their studies, the BDI was administered in a separate session and they only invited participation from individuals with no depression or at least moderate depression (i.e., they excluded people with mild depression). Second, participants completed the manipulation and all other measures at an individual experimental session (not in groups). Third, different experimenters ran the different parts of the study to make believable the “different studies” aspect of the cover story. Fourth, participants completed the experimental manipulation and all measures on paper and pencil rather than on the computer. That said, we believe the current study remains an accurate replication of this work given that we obtained all original materials from Lyubomirsky and do not believe our small changes in administration protocol should affect the results. Furthermore, by including participants with mild depression in our analyses and exploring depression continuously, we are capturing additional information while still representing the comparisons made in the original studies.

Results

Overview of Analyses

Lyubomirsky, Nolen-Hoeksema, and colleagues primarily conducted planned contrasts with depression dichotomized (depressed and non-depressed). Setting our gratitude condition aside, we reported these same analyses in Supplemental Material. In the main manuscript, we decided to model depression scores as continuous in regression-based analyses to align with literature supporting depression as a dimensional construct (e.g., Markon et al., 2011) and to increase variance across participants while improving statistical power, validity, and reliability (Cohen, 1983; DeCoster et al., 2009; Markon et al., 2011). Specifically, we used the multicategorical function in Hayes’ (2018) PROCESS macro (version 3.3, Model 1) for SPSS to calculate the unstandardized regression coefficients in separate multiple regression models predicting depressed mood, negative affect, and positive affect with the following predictors: gratitude (vs. rumination), distraction (vs. rumination), continuous baseline depression scores, gratitude (vs. rumination) by baseline depression interaction, distraction (vs. rumination) by baseline depression interaction, and baseline levels of the dependent variables (see results in Table 2 and Figure 1). The gratitude and distraction conditions are dummy coded and the rumination condition is the reference group. The distraction (vs rumination) by baseline depression interaction addresses Hypotheses 1a (on depressed mood) and 1b (on negative affect), and the gratitude (vs. rumination) predictor addresses Hypotheses 2a (on depressed mood), 2b (on negative affect), and 2c (on positive affect). We also ran analyses with the distraction condition as the reference group to compare it directly to the gratitude condition (to test Hypothesis 3 on positive affect). Moreover, we used these regression models to test our exploratory hypotheses. For example, although we did not make an a priori hypothesis about whether there would be a gratitude (vs. rumination) by baseline depression interaction on depressed mood, negative affect, or positive affect, we included that predictor for exploratory purposes. Similarly, although we did not make an a priori hypothesis about potential differences between the gratitude and distraction condition on depressed mood and negative affect, we ran that comparison to inform future work. In addition, when the interaction between condition and depression was significant (p < .05), we used the Johnson-Neyman regions of significance technique to report at what baseline depression score the effect of condition becomes significant (Hayes, 2018).

Table 2.

Unstandardized Regression Coefficients (Standard Errors) Predicting Affect from Condition, Baseline Depression, and Condition by Baseline Depression

Effects	Depressed Mood		Negative Affect		Positive Affect
	b (SE)	95% CI	b (SE)	95% CI	b (SE)	95% CI
Study 1
Constant	2.02 (0.08)***	[1.87, 2.16]	2.16 (0.07)***	[2.02, 2.30]	3.64 (0.09)***	[3.46, 3.82]
Pre-Test	0.66 (0.04)***	[0.57, 0.74]	0.69 (0.05)***	[0.59, 0.78]	0.84 (0.04)***	[0.77, 0.92]
Gratitude (vs. Rumination)	−0.22 (0.11)*	[−0.43, −0.01]	−0.26 (0.10)**	[−0.46, −0.07]	0.53 (0.13)***	[0.28, 0.78]
Distraction (vs. Rumination)	−0.20 (0.11)†	[−0.40, 0.01]	−0.33 (0.10)**	[−0.53, −0.13]	−0.22 (0.13)†	[−0.48, 0.03]
Depression (Centered)	0.03 (0.01)*	[0.01, 0.05]	0.02 (0.01)	[−0.005, 0.04]	−0.01 (0.01)	[−0.04, 0.01]
Gratitude (vs. Rumination) × Depression	−0.02 (0.01)	[−0.04, 0.01]	−0.003 (0.01)	[−0.03, 0.02]	0.01 (0.02)	[−0.02, 0.05]
Distraction (vs. Rumination) × Depression	−0.02 (0.01)	[−0.04, 0.01]	−0.02 (0.01)	[−0.04, 0.01]	0.01 (0.01)	[−0.02, 0.04]
Gratitude (vs. Distraction)	−0.03 (0.11)	[−0.24, 0.18]	0.06 (0.10)	[−0.14, 0.26]	0.76 (0.13)***	[0.50, 1.01]
Gratitude (vs. Distraction) × Depression	0.003 (0.01)	[−0.02, 0.03]	0.01 (0.01)	[−0.01, 0.04]	0.003 (0.01)	[−0.03, 0.03]
Study 2
Constant	2.21 (0.09)***	[2.04, 2.38]	2.18 (0.07)***	[2.04, 2.33]	3.56 (0.10)***	[3.35, 3.76]
Pre-Test	0.80 (0.05)***	[0.71, 0.90]	0.72 (0.05)***	[0.62, 0.82]	0.84 (0.04)***	[0.76, 0.93]
Gratitude (vs. Rumination)	−0.22 (0.12)†	[−0.46, 0.02]	−0.31 (0.10)**	[−0.51, −0.10]	0.55 (0.15)***	[0.26, 0.84]
Distraction (vs. Rumination)	−0.38 (0.12)**	[−0.62, −0.14]	−0.23 (0.10)*	[−0.44, −0.03]	−0.07 (0.15)	[−0.36, 0.21]
Depression (Centered)	0.02 (0.01)	[−0.01, 0.04]	0.03 (0.01)**	[0.01, 0.05]	0.01 (0.01)	[−0.02, 0.03]
Gratitude (vs. Rumination) × Depression	−0.03 (0.02)†	[−0.06, 0.004]	−0.03 (0.01)*	[−0.05, −0.001]	−0.01 (0.02)	[−0.01, 0.05]
Distraction (vs. Rumination) × Depression	−0.03 (0.01)*	[−0.06, −.005]	−0.03 (0.01)*	[−0.05, −0.003]	0.02 (0.02)	[−0.01, 0.05]
Gratitude (vs. Distraction)	0.16 (0.12)	[−0.08, 0.40]	−0.08 (0.11)	[−0.28, 0.13]	0.63 (0.15)***	[0.33, 0.92]
Gratitude (vs. Distraction) × Depression	0.01 (0.01)	[−0.02, 0.04]	0.001 (0.01)	[−0.02, 0.03]	−0.03 (0.02)†	[−0.07, 0.001]
Study 3
Constant	2.21 (0.06)***	[2.10, 2.32]	2.12 (0.04)***	[2.03, 2.21]	3.60 (0.06)***	[3.49, 3.71]
Pre-Test	0.80 (0.03)***	[0.75, 0.85]	0.81 (0.02)***	[0.77, 0.86]	0.94 (0.02)***	[0.90, 0.98]
Gratitude (vs. Rumination)	−0.12 (0.08)	[−0.27, 0.03]	−0.11 (0.06)†	[−0.23, 0.01]	0.20 (0.08)*	[0.05, 0.35]
Distraction (vs. Rumination)	−0.24 (0.08)**	[−0.40, −0.09]	−0.11 (0.06)†	[−0.23, 0.02]	0.09 (0.08)	[−0.06, 0.25]
Depression (Centered)	0.04 (0.01)***	[0.02, 0.06]	0.02 (0.01)**	[0.01, 0.04]	−0.003 (0.01)	[−0.02, 0.02]
Gratitude (vs. Rumination) × Depression	−0.04 (0.01)**	[−0.07, −0.02]	−0.01 (0.01)	[−0.04, 0.01]	0.02 (0.01)	[−0.01, 0.04]
Distraction (vs. Rumination) ×Depression	−0.04 (0.01)**	[−0.07, −0.01]	−0.01 (0.01)	[−0.03, 0.01]	0.01 (0.01)	[−0.01, 0.04]
Gratitude (vs. Distraction)	0.13 (0.08)†	[−0.02, 0.28]	−0.003 (0.06)	[−0.12, 0.12]	0.10 (0.08)	[−0.05, 0.26]
Gratitude (vs. Distraction) × Depression	0.0001 (0.01)	[−0.02, 0.02]	−0.004 (0.01)	[−0.02, 0.02]	0.004 (0.01)	[−0.02, 0.03]
Study 4
Constant	2.14 (0.07)***	[1.99, 2.29]	2.03 (0.06)***	[1.92, 2.14]	3.27 (0.10)***	[3.08, 3.46]
Pre-Test	0.67 (0.04)***	[−0.58, 0.75]	0.58 (0.04)***	[0.51, 0.66]	0.91 (0.05)***	[0.81, 1.01]
Gratitude (vs. Rumination)	−0.32 (0.11)**	[−0.53, −0.11]	−0.38 (0.08)***	[−0.54, −0.22]	0.54 (0.14)***	[0.27, 0.81]
Distraction (vs. Rumination)	−0.46 (0.11)***	[−0.67, −0.25]	−0.25 (0.08)**	[−0.41, −0.09]	0.07 (0.14)	[−0.20, 0.35]
Depression (Centered)	0.03 (0.01)**	[0.01, 0.06]	0.03 (0.01)***	[0.02, 0.05]	−0.002 (0.01)	[−0.03, 0.02]
Gratitude (vs. Rumination) × Depression	−0.03 (0.01)*	[−0.05, −0.001]	−0.04 (0.01)***	[−0.06, −0.02]	0.003 (0.02)	[−0.03, 0.04]
Distraction (vs. Rumination) × Depression	−0.04 (0.01)**	[−0.07, −0.01]	−0.03 (0.01)*	[−0.05, −0.004]	0.004 (0.02)	[−0.03, 0.04]
Gratitude (vs. Distraction)	0.14 (0.11)	[−0.07, 0.35]	−0.13 (0.08)	[−0.29, 0.03]	0.47 (0.14)***	[0.19, 0.74]
Gratitude (vs. Distraction) × Depression	0.01 (0.01)	[−0.02, 0.04]	−0.01 (0.01)	[−0.03, 0.01]	−0.002 (0.02)	[−0.04, 0.03]
Study 5
Constant	1.98 (0.07)***	[1.85, 2.10]	1.93 (0.05)***	[1.84, 2.02]	3.35 (0.07)***	[3.22, 3.48]
Pre-Test	0.83 (0.05)***	[0.74, 0.92]	0.82 (0.04)***	[0.75, 0.89]	0.92 (0.03)***	[0.87, 0.98]
Gratitude (vs. Rumination)	−0.15 (0.09)	[−0.34, 0.03]	−0.27 (0.07)***	[−0.40, −0.13]	0.57 (0.09)***	[0.38, 0.75]
Distraction (vs. Rumination)	−0.12 (0.09)	[−0.30, 0.06]	−0.21 (0.07)**	[−0.34, −0.08]	−0.04 (0.09)	[−0.23, 0.14]
Depression (Centered)	0.01 (0.01)	[−0.01, 0.03]	0.01 (0.005)	[−0.003, 0.01]	0.01 (0.01)	[−0.005, 0.02]
Gratitude (vs. Rumination) × Depression	0.01 (0.01)	[−0.01, 0.02]	−0.01 (0.01)	[−0.02, 0.003]	−0.02 (0.01)†	[−0.03, 0.002]
Distraction (vs. Rumination) × Depression	0.002 (0.01)	[−0.01, 0.02]	−0.01 (0.01)	[−0.02, 0.002]	−0.0005 (0.01)	[−0.02, 0.02]
Gratitude (vs. Distraction)	−0.04 (0.09)	[−0.22, 0.15]	−0.06 (0.07)	[−0.19, 0.08]	0.61 (0.09)***	[0.42, 0.79]
Gratitude (vs. Distraction) × Depression	−0.007 (0.01)	[−0.01, 0.02]	0.0004 (0.01)	[−0.01, 0.01]	−0.02 (0.01)†	[−0.03, 0.003]

Note. Under each study, the first set of effects includes a model in which condition is dummy coded with the rumination condition as the reference group (0) and the gratitude and distraction conditions coded as 1, baseline depression is centered, and baseline levels of each outcome are centered. Thus, the constant is the average score on the dependent variable for people in the rumination condition at the average level of baseline depression and the baseline outcome. Gratitude (vs. Rumination) is the effect of being in the gratitude (vs. rumination) condition in addition to the constant. Distraction (vs. Rumination) is the effect of being in the distraction (vs. rumination) condition in addition to the constant. Centered baseline depression is the additional effect of being above or below the average level of baseline depression for people in the rumination condition. Gratitude (vs. Rumination) × Depression is the additional effect of being above or below the average level of baseline depression for participants in the gratitude condition (beyond the effect of baseline depression among those in the rumination condition). Distraction (vs. Rumination) × Depression is the additional effect of being above or below the average level of baseline depression for participants in the distraction condition (beyond the effect of baseline depression among those in the rumination condition). After the first set of effects under each study, followed by a space, the Gratitude (vs. Distraction) and the Gratitude (vs. Distraction) × Depression effects are from a second model in which the distraction condition is the reference group (0) and the gratitude condition is coded as a 1. The Gratitude (vs. Distraction) effect represents the effect of being in the gratitude (vs. distraction) condition for people at the average level of depression. The Gratitude (vs. Distraction) × Depression effect is the additional effect of being above or below the average level of baseline depression for participants in the gratitude condition (beyond the effect of baseline depression among those in the distraction condition). Confidence intervals were calculated with Ordinary Least Squares regression (i.e., they are not bootstrapped).

^†p < .10

^*p < .05

^**p < .01

^***p < .001

Figure 1.

Model-Predicted Affect by Condition and Baseline Depression Across Studies

Note. The five rows present model-predicted means from a multiple linear regression model predicting depressed mood, negative affect, and positive affect (from left to right, respectively) from experimental condition, centered baseline depression (as measured by the Beck Depression Inventory-II), and the interactions between condition and centered baseline depression, controlling for baseline levels of the respective outcome variable. From top to bottom, the five panels represent results from each of our five studies. For baseline depression, low is −1 SD, medium is the mean, and high is +1 SD, except in Study 3 and Study 5 in which cases −1 SD was not an observed value, so low is the lowest depression score observed, which was 0 in both studies. Error bars represent standard errors. Predicted means and standard errors were generated by the plot=2, moments=1 command in Model 1 of PROCESS version 3.3 (Hayes, 2018).

To get a global picture of our affect analyses across the five studies, we conducted meta-analyses on depressed mood (Table 3a), negative affect (Table 3b), and positive affect (Table 3c). Specifically, per recommendations by Aloe and Becker (2012), we calculated semi-partial correlations for each predictor in each study and then calculated the fixed and random effects across the five studies by predictor using metacor within the meta package (version 5.2–0; Schwarzer, 2022; Schwarzer et al., 2015) in R version 4.1.3 (2022–03-10). Semi-partial correlations are conceptually closest to multiple regression coefficients as they represent the unique explanatory power of each predictor after controlling for the other predictors. Because we anticipated potential between-study heterogeneity between our online and student samples, we calculated random effects models (in addition to the fixed effects models, which assumes homogeneity across studies). Specifically, we used the restricted maximum likelihood estimator to calculate τ² (Veroniki et al., 2016; Viechtbauer, 2005), the Q-profile method to calculate confidence intervals around τ², and also applied a Knapp-Hartung adjustment (Knapp & Hartung, 2003). Indices of between-study heterogeneity (confidence intervals around τ², Q [Cochran, 1954; Hoaglin, 2016], and I² [Higgins & Thompson, 2002]) demonstrated minimal between-study variability on our predicted effects, so we did not move forward to explore sample type (online vs. student) as a moderator. We consulted Harrer et al. (2021) for guidance in employing these meta-analytic procedures.

Table 3a.

Meta-Analyzed Semi-Partial Correlations Between Predictors and Depressed Mood

Effect	k	N	r sp	95% CI around rsp	t-statistic	τ2	95% CI around τ2	Q	I 2
Fixed Effects
Pre-Test	5	1950	.56***	[.53, .59]	27.85
Gratitude (vs. Rumination)	5	1950	−.06*	[−.10, −.01]	−2.47
Distraction (vs. Rumination)	5	1950	−.08***	[−.12, −.03]	−3.36
Depression (Centered)	5	1950	.07**	[.02, .11]	2.90
Gratitude (vs. Rumination) × Depression	5	1950	−.04†	[−.08, .01]	−1.70
Distraction (vs. Rumination) × Depression	5	1950	−.05*	[−.10, −.01]	−2.28
Gratitude (vs. Distraction)	5	1950	.02	[−.02, .07]	0.90
Gratitude (vs. Distraction) × Depression	5	1950	.01	[−.03, .05]	0.56
Random Effects
Pre-Test	5	1950	.56***	[.48, .63]	16.69	0.004	[0.00, 0.06]	10.69*	62.60
Gratitude (vs. Rumination)	5	1950	−.06*	[−.09, −.02]	−4.33	0.00	[0.00, 0.004]	1.30	0.00
Distraction (vs. Rumination)	5	1950	−.08*	[−.13, −.02]	−3.84	0.00	[0.00, 0.02]	3.06	0.00
Depression (Centered)	5	1950	.07**	[.03, .11]	4.64	0.00	[0.00, 0.01]	1.56	0.00
Gratitude (vs. Rumination) × Depression	5	1950	−.04	[−.09, .02]	−1.94	0.00	[0.00, 0.01]	3.08	0.00
Distraction (vs. Rumination) × Depression	5	1950	−.05*	[−.10, −.001]	−2.85	0.00	[0.00, 0.01]	2.57	0.00
Gratitude (vs. Distraction)	5	1950	.02	[−.01, .06]	1.64	0.00	[0.00, 0.004]	1.21	0.00
Gratitude (vs. Distraction) × Depression	5	1950	.01†	[−.002, .02]	2.38	0.00	[0.00, 0.00]	0.22	0.00

Note. See Note under Table 2 for regression coding scheme. Per recommendations by Aloe and Becker (2012) we calculated semi-partial correlations for each predictor in each study and then calculated the fixed and random effects across the five studies by predictor using metacor within the meta package (inverse variance weighting is used for pooling; version 5.2–0; Schwarzer, 2022; Schwarzer et al., 2015) in R, version 4.1.3 (2022–03-10). In the random effects models, we used the restricted maximum likelihood estimator to calculate τ² (Veroniki et al., 2016; Viechtbauer, 2005), the Q-profile method to calculate confidence intervals around τ², and also applied a Knapp-Hartung adjustment (Knapp & Hartung, 2003). Indices of between-study heterogeneity (confidence intervals around τ², Q [Cochran, 1954; Hoaglin, 2016], and I² [Higgins & Thompson, 2002]) demonstrated minimal between-study variability on our predicted effects, so we did not move forward to explore our anticipated moderator (online vs. student sample). We consulted Harrer et al. (2021) for guidance in employing these meta-analytic procedures.

^†p < .10

^*p < .05

^**p < .01

^***p < .001

Table 3b.

Meta-Analyzed Semi-Partial Correlations Between Predictors and Negative Affect

Effect	k	N	r sp	95% CI around rsp	t-statistic	τ2	95% CI around τ2	Q	I 2
Fixed Effects
Pre-Test	5	1950	.59***	[.56, .62]	29.91
Gratitude (vs. Rumination)	5	1950	−.09***	[−.13, −.05]	−3.96
Distraction (vs. Rumination)	5	1950	−.08***	[−.12, −.03]	−3.35
Depression (Centered)	5	1950	.07**	[.03, .12]	3.23
Gratitude (vs. Rumination) × Depression	5	1950	−.05*	[−.09, −.005]	−2.16
Distraction (vs. Rumination) × Depression	5	1950	−.05*	[−.09, −.003]	−2.11
Gratitude (vs. Distraction)	5	1950	−.01	[−.06, .03]	−0.61
Gratitude (vs. Distraction) × Depression	5	1950	−.002	[−.05, 0.04]	−0.08
Random Effects
Pre-Test	5	1950	0.58***	[.53, .64]	21.63	0.003	[0.00, 0.03]	8.10†	50.60
Gratitude (vs. Rumination)	5	1950	−.09*	[−.16, −.03]	−4.01	0.0005	[0.00, 0.02]	4.20	4.80
Distraction (vs. Rumination)	5	1950	−.08*	[−.12, −.03]	−4.36	0.0000	[0.00, 0.008]	2.36	0.00
Depression (Centered)	5	1950	.07*	[.01, .14]	3.23	0.0001	[0.00, 0.02]	4.00	0.00
Gratitude (vs. Rumination) × Depression	5	1950	−.05†	[−.11, .01]	−2.38	0.0000	[0.00, 0.02]	3.32	0.00
Distraction (vs. Rumination) × Depression	5	1950	−.05*	[−.09, −.01]	−3.49	0.0000	[0.00, 0.005]	1.46	0.00
Gratitude (vs. Distraction)	5	1950	−.01	[−.05, 0.02]	−1.17	0.0000	[0.00, 0.004]	1.10	0.00
Gratitude (vs. Distraction) × Depression	5	1950	−.002	[−.03, 0.03]	−0.16	0.0000	[0.00, 0.004]	1.07	0.00

Note. See note under Table 3a for information on how these meta-analyses were conducted. Model-Predicted Affect by

Table 3c.

Meta-Analyzed Semi-Partial Correlations Between Predictors and Positive Affect

Effect	k	N	r sp	95% CI around rsp	t-statistic	τ2	95% CI around τ2	Q	I 2
Fixed Effects
Pre-Test	5	1950	.79***	[.77, .80]	46.82
Gratitude (vs. Rumination)	5	1950	.11***	[.06, .15]	4.76
Distraction (vs. Rumination)	5	1950	−.004	[−.05, .04]	−0.18
Depression (Centered)	5	1950	.001	[−.04, .05]	0.05
Gratitude (vs. Rumination) × Depression	5	1950	−.001	[−.05, .04]	−0.03
Distraction (vs. Rumination) × Depression	5	1950	.02	[−.03, .06]	0.74
Gratitude (vs. Distraction)	5	1950	.11***	[.07, .16]	4.94
Gratitude (vs. Distraction) × Depression	5	1950	−.02	[−.06, .03]	−0.73
Random Effects
Pre-Test	5	1950	.77***	[.68, .84]	14.29	0.02	[0.01, 0.20]	46.29***	91.40
Gratitude (vs. Rumination)	5	1950	.11**	[.05, .17]	5.10	0.001	[0.00, 0.01]	3.82	0.00
Distraction (vs. Rumination)	5	1950	−.004	[−.04, .03]	−0.30	0.00	[0.00, 0.01]	1.49	0.00
Depression (Centered)	5	1950	.001	[−.03, .03]	0.12	0.00	[0.00, 0.002]	0.77	0.00
Gratitude (vs. Rumination) × Depression	5	1950	−.001	[−.04, .04]	−0.04	0.00	[0.00, 0.005]	1.45	0.00
Distraction (vs. Rumination) × Depression	5	1950	.02†	[−.005, .04]	2.18	0.00	[0.00, 0.00]	0.46	0.00
Gratitude (vs. Distraction)	5	1950	.12*	[.04, .21]	3.97	0.003	[0.00, 0.03]	8.24†	51.40
Gratitude (vs. Distraction) × Depression	5	1950	−.02	[−.06, .02]	−1.18	0.00	[0.00, 0.01]	1.52	0.00

Note. See note under Table 3a for information on how these meta-analyses were conducted.

Next, because we expected that condition-induced changes in positive and negative affect would predict changes in our thought pattern outcomes (e.g., the Cognitive Bias Questionnaire and the Twenty-Statements Test), we also used Hayes’ PROCESS macro to assess the indirect effect of condition on our thought pattern outcomes via positive and negative affect in a parallel mediation model, while adjusting for baseline levels of positive and negative affect (Model 4 in PROCESS version 3.3; Hayes, 2018). For each effect, we provided the estimated indirect effect (ab) via post-test positive and negative affect (separately), as well as a 95% bootstrap confidence interval based on 5,000 bootstrap samples around the indirect effect in Tables 4– 8 (Preacher et al., 2007). Confidence intervals that do not include zero are considered statistically significant. This non-parametric resampling technique maximizes power while maintaining Type I error rate (Shrout & Bolger, 2002). We also included unstandardized parameter estimates and 95% confidence intervals for the a, b, and c’ paths from Ordinary Least Squares regression analyses.

Table 4.

Indirect Effect of Condition on Twenty Statements Test via Positive and Negative Affect (Study 1 and Study 3)

	Gratitude vs. Rumination		Distraction vs. Rumination		Gratitude vs. Distraction
	Effect (SE)	95% CI	Effect (SE)	95% CI	Effect (SE)	95% CI
STUDY 1
Number of Responses
Positive Affect
a	0.53 (0.13)***	[0.28, 0.78]	−0.22 (0.13)†	[−0.47, 0.04]	0.74 (0.13)***	[0.49, 1.00]
b	−0.47 (0.36)	[−1.18, 0.24]	−0.47 (0.36)	[−1.18, 0.24]	−0.47 (0.36)	[−1.18, 0.24]
Indirect Effect (ab)	−0.25 (0.23)	[−0.74, 0.16]	0.10 (0.11)	[−0.08, 0.38]	−0.35 (0.30)	[−0.96, 0.20]
Negative Affect
a	−0.24 (0.10)*	[−0.44, −0.04]	−0.33 (0.10)**	[−0.53, −0.13]	0.09 (0.10)	[−0.11, 0.29]
b	0.38 (0.46)	[−0.51, 1.28]	0.38 (0.46)	[−0.51, 1.28]	0.38 (0.46)	[−0.51, 1.28]
Indirect Effect (ab)	−0.09 (0.15)	[−0.44, 0.16]	−0.13 (0.19)	[−0.53, 0.23]	0.03 (0.07)	[−0.10, 0.21]
Direct Effect (c’)	0.73 (0.81)	[−0.87, 2.33]	0.29 (0.81)	[−1.30, 1.89]	0.44 (0.84)	[−1.21, 2.09]
Coder-Rated Positivity
Positive Affect
a	0.53 (0.13)***	[0.28, 0.78]	−0.22 (0.13)†	[−0.47, 0.04]	0.74 (0.13)***	[0.49, 1.00]
b	0.24 (0.09)**	[0.07, 0.41]	0.24 (0.09)**	[0.07, 0.41]	0.24 (0.09)**	[0.07, 0.41]
Indirect Effect (ab)	0.13 (0.06)	[0.03, 0.25]	−0.05 (0.04)	[−0.14, 0.01]	0.18 (0.07)	[0.05, 0.33]
Negative Affect
a	−0.24 (0.10)*	[−0.44, −0.04]	−0.33 (0.10)**	[−0.53, −0.13]	0.09 (0.10)	[−0.11, 0.29]
b	−0.23 (0.11)*	[−0.45, −0.02]	−0.23 (0.11)*	[−0.45, −0.02]	−0.23 (0.11)*	[−0.45, −0.02]
Indirect Effect (ab)	0.06 (0.03)	[0.003, 0.14]	0.08 (0.04)	[0.01, 0.17]	−0.02 (0.03)	[−0.08, 0.02]
Direct Effect (c’)	0.88 (0.19)***	[0.50, 1.26]	0.04 (0.19)	[−0.34, 0.42]	0.84 (0.20)***	[0.44, 1.23]
Coder-Rated Negativity
Positive Affect
a	0.53 (0.13)***	[0.28, 0.78]	−0.22 (0.13)†	[−0.47, 0.04]	0.74 (0.13)***	[0.49, 1.00]
b	−0.09 (0.05)†	[−0.19, 0.01]	−0.09 (0.05)†	[−0.19, 0.01]	−0.09 (0.05)†	[−0.19, 0.01]
Indirect Effect (ab)	−0.05 (0.03)	[−0.11, −0.004]	0.02 (0.02)	[−0.003, 0.06]	−0.07 (0.04)	[−0.15, −0.004]
Negative Affect
a	−0.24 (0.10)*	[−0.44, −0.04]	−0.33 (0.10)**	[−0.53, −0.13]	0.09 (0.10)	[−0.11, 0.29]
b	0.43 (0.06)***	[0.30, 0.56]	0.43 (0.06)***	[0.30, 0.56]	0.43 (0.06)***	[0.30, 0.56]
Indirect Effect (ab)	−0.11 (0.06)	[−0.24, −0.02]	−0.14 (0.06)	[−0.29, −0.04]	0.04 (0.04)	[−0.04, 0.13]
Direct Effect (c’)	−0.17 (0.11)	[−0.39, 0.06]	−0.25 (0.11)*	[−0.47, −0.02]	0.08 (0.12)	[−0.15, 0.31]
STUDY 3
Number of Responses
Positive Affect
a	0.21 (0.08)**	[0.06, 0.36]	0.09 (0.08)	[−0.07, 0.25]	0.12 (0.08)	[−0.03, 0.27]
b	0.67 (0.26)*	[0.16, 1.19]	0.67 (0.26)*	[0.16, 1.19]	0.67 (0.26)*	[0.16, 1.19]
Indirect Effect (ab)	0.14 (0.08)	[0.02, 0.33]	0.06 (0.06)	[−0.03, 0.19]	0.08 (0.07)	[−0.02, 0.23]
Negative Affect
a	−0.09 (0.06)	[−0.21, 0.03]	−0.10 (0.06)	[−0.22, 0.03]	0.01 (0.06)	[−0.11, 0.13]
b	−0.04 (0.33)	[−0.68, 0.61]	−0.04 (0.33)	[−0.68, 0.61]	−0.04 (0.33)	[−0.68, 0.61]
Indirect Effect (ab)	0.003 (0.04)	[−0.08, 0.09]	0.004 (0.04)	[−0.10, 0.09]	−0.0003 (0.03)	[−0.04, 0.06]
Direct Effect (c’)	0.18 (0.51)	[−0.83, 1.18]	0.23 (0.52)	[−0.79, 1.25]	−0.06 (0.50)	[−1.04, 0.93]
Coder-Rated Positivity
Positive Affect
a	0.23 (0.08)**	[0.07, 0.38]	0.11 (0.08)	[−0.05, 0.27]	0.12 (0.08)	[−0.04, 0.27]
b	0.57 (0.09)***	[0.41, 0.74]	0.57 (0.09)***	[0.41, 0.74]	0.57 (0.09)***	[0.41, 0.74]
Indirect Effect (ab)	0.13 (0.05)	[0.04, 0.24]	0.06 (0.04)	[−0.02, 0.15]	0.07 (0.05)	[−0.02, 0.17]
Negative Affect
a	−0.08 (0.06)	[−0.20, 0.05]	−0.10 (0.06)	[−0.23, 0.03]	0.02 (0.06)	[−0.10, 0.15]
b	−0.22 (0.11)*	[−0.43, −0.02]	−0.22 (0.11)*	[−0.43, −0.02]	−0.22 (0.11)*	[−0.43, −0.02]
Indirect Effect (ab)	0.02 (0.02)	[−0.01, 0.06]	0.02 (0.02)	[−0.01, 0.07]	−0.01 (0.02)	[−0.04, 0.03]
Direct Effect (c’)	0.35 (0.16)*	[0.03, 0.66]	0.03 (0.16)	[−0.29, 0.35]	0.32 (0.16)*	[0.004, 0.63]
Coder-Rated Negativity
Positive Affect
a	0.23 (0.08)**	[0.07, 0.38]	0.11 (0.08)	[−0.05, 0.27]	0.12 (0.08)	[−0.04, 0.27]
b	−0.14 (0.05)**	[−0.25, −0.04]	−0.14 (0.05)**	[−0.25, −0.04]	−0.14 (0.05)**	[−0.25, −0.04]
Indirect Effect (ab)	−0.03 (0.02)	[−0.08, −0.01]	−0.02 (0.01)	[−0.04, 0.01]	−0.02 (0.01)	[−0.05, 0.01]
Negative Affect
a	−0.08 (0.06)	[−0.20, 0.05]	−0.10 (0.06)	[−0.23, 0.03]	0.02 (0.06)	[−0.03, 0.23]
b	0.20 (0.07)**	[0.08, 0.33]	0.20 (0.07)**	[0.08, 0.33]	0.20 (0.07)**	[0.08, 0.33]
Indirect Effect (ab)	−0.02 (0.01)	[−0.05, 0.01]	−0.02 (0.02)	[−0.06, 0.01]	0.005 (0.01)	[−0.02, 0.04]
Direct Effect (c’)	−0.08 (0.10)	[−0.28, 0.12]	−0.26 (0.10)**	[−0.46, −0.06]	0.18 (0.10)†	[−0.01, 0.38]

Note. Condition is dummy-codedratitude vs. Distractionnppear once, so not as big of a deal to compare??es in the main manuscript? be compared directly to dist. For the first two columns, Rumination is the reference group and Gratitude and Distraction were entered as separate predictors in the same model. The b is the same in all three columns (Gratitude vs. Rumination, Distraction vs. Rumination, and Gratitude vs. Distraction) because the first two columns are from the same model and the third column is from a model in which the condition composition is the same (only the referent group has been changed). For the third column (Gratitude vs. Distraction), Distraction became the reference group and Gratitude and Rumination were entered as separate predictors in the same model. The Rumination vs. Distraction information was redundant with the first model, so is not included here. Standard errors and 95% confidence intervals for the indirect effect were calculated with the percentile bootstrap approach based on 5,000 bootstrap samples (Hayes, 2018). The unstandardized parameter estimates and 95% confidence intervals for the a, b, and c’ paths were calculated with Ordinary Least Squares regression. For indirect effects analyses, confidence intervals are considered significant if they do not include zero and are designated in bold. For a, b, and c’ paths, we used the following schema to indicate significance:

^†p < .10

^*p < .05

^**p < .01

^***p < .001

Table 8.

Indirect Effect of Condition on Judgment of Pleasant Activities via Positive and Negative Affect (Study 4)

	Gratitude vs. Rumination		Distraction vs. Rumination		Gratitude vs. Distraction
	Effect (SE)	95% CI	Effect (SE)	95% CI	Effect (SE)	95% CI
Enjoyment of Pleasant Activities
Positive Affect
a	0.52 (0.14)***	[0.25, 0.80]	0.07 (0.14)	[−0.20, 0.34]	0.45 (0.14)**	[0.18, 0.72]
b	0.09 (0.02)***	[0.04, 0.14]	0.09 (0.02)***	[0.04, 0.14]	0.09 (0.02)***	[0.04, 0.14]
Indirect Effect (ab)	0.05 (0.02)	[0.02, 0.09]	0.01 (0.01)	[−0.02, 0.03]	0.04 (0.02)	[0.01, 0.08]
Negative Affect
a	−0.38 (0.08)***	[−0.55, −0.22]	−0.23 (0.08)**	[−0.40, −0.07]	−0.15 (0.08)†	[−0.32, 0.02]
b	−0.04 (0.04)	[−0.12, 0.04]	−0.04 (0.04)	[−0.12, 0.04]	−0.04 (0.04)	[−0.12, 0.04]
Indirect Effect (ab)	0.02 (0.02)	[−0.02, 0.05]	0.01 (0.01)	[−0.02, 0.03]	0.01 (0.01)	[−0.01, 0.03]
Direct Effect (c’)	−0.07 (0.06)	[−0.19, 0.05]	−0.05 (0.06)	[−0.17, 0.06]	−0.02 (0.06)	[−0.14, 0.09]
Likelihood of Engagement in Pleasant Activities
Positive Affect
a	0.52 (0.14)***	[0.25, 0.80]	0.07 (0.14)	[−0.20, 0.34]	0.45 (0.14)**	[0.18, 0.72]
b	0.08 (0.03)**	[0.02, 0.14]	0.08 (0.03)**	[0.02, 0.14]	0.08 (0.03)**	[0.02, 0.14]
Indirect Effect (ab)	0.04 (0.02)	[0.01, 0.09]	0.01 (0.01)	[−0.02, 0.03]	0.04 (0.02)	[0.01, 0.08]
Negative Affect
a	−0.38 (0.08)***	[−0.55, −0.22]	−0.23 (0.08)**	[−0.40, −0.07]	−0.15 (0.08) †	[−0.32, 0.02]
b	−0.003 (0.05)	[−0.10, 0.09]	−0.003 (0.05)	[−0.10, 0.09]	−0.003 (0.05)	[−0.10, 0.09]
Indirect Effect (ab)	0.001 (0.02)	[−0.04, 0.05]	0.001 (0.01)	[−0.03, 0.03]	0.001 (0.01)	[−0.02, 0.02]
Direct Effect (c’)	−0.005 (0.07)	[−0.15, 0.14]	−0.02 (0.07)	[−0.16, 0.11]	0.02 (0.07)	[−0.12, 0.16]

Note. See note in Table 4 for coding scheme. For indirect effects analyses, confidence intervals are considered significant if they do not include zero and are designated in bold.

Specifically, we tested indirect effects on twelve thinking pattern outcomes: three from the Twenty Statements Test (number of responses, coder-rated positivity of thought-action repertoires, coder-rated negativity of thought-action repertoires; Studies 1 and 3), two from the Cognitive Bias Questionnaire (depressed/distorted and nondepressed/nondistorted thinking; Studies 1 and 5), two from the Event Frequency Task (positive and negative event frequency; Studies 2 and 3), three from the Means-Ends Problem Solving Task (global effectiveness in solving Story 1, global effectiveness in solving Story 2, and percentage of effective solutions across four stories; Study 2), and two from the Judgment of Pleasant Activities measure (enjoyability and likelihood of engaging; Study 4). For each thought pattern outcome, we tested six possible indirect effects (only four of which were hypothesized to be significant): (1) distraction (vs. rumination) via negative affect (included under Hypothesis 4a), (2) distraction (vs. rumination) via positive affect (no hypothesis), (3) gratitude (vs. rumination) via negative affect (included under Hypothesis 4a), (4) gratitude (vs. rumination) via positive affect (included under Hypothesis 4b), (5) gratitude (vs. distraction) via negative affect (no hypothesis), and (6) gratitude (vs. distraction) via positive affect (included under Hypothesis 4b). Just like with our primary affect analyses, the distraction (vs. rumination) and gratitude (vs. rumination) indirect effects were included in one model (with rumination as the reference group), whereas the gratitude (vs. distraction) comparison was included in a separate model (with distraction as the reference group and rumination included as a predictor). Although we did not hypothesize an indirect effect of gratitude (vs. distraction) via negative affect (for example), we had to include it in the model to explore the unique indirect effect via positive affect, above and beyond negative affect. We decided to use the negative affect composite as a mediator instead of depressed mood because it represents a broader range of negative emotions. As such, we did not include depressed mood as a mediator in these analyses because of the large correlation between it and negative affect.

In addition, although we expected some interactions between the distraction (vs. rumination) condition and baseline depression scores on negative affect, we found no significant indices of moderated mediation in any of the studies (using Model 7 of PROCESS version 3.3; Hayes, 2018), so we present the indirect effect analyses unmoderated by baseline depression.

Lastly, in all studies, Lyubomirsky, Nolen-Hoeksema, and colleagues reported that they explored the potential effects of gender on the dependent variables. In some cases, they found a main effect of gender, but never an interaction with condition or baseline depression and therefore they did not report gender as part of their main analyses. Importantly, we found some interactions between condition, baseline depression, and gender, but they were inconsistent across studies and outcomes. We provide these analyses in Supplemental Material for full transparency.

Baseline Analyses

We found no significant differences among conditions on continuous baseline depression scores (Fs < 1.40), baseline depressed mood (Fs < 1.22), baseline negative affect (Fs < 2.28), or baseline positive affect (Fs < 2.28) in any of the studies. See pre- and post-test means and standard deviations by condition and Study in Table 1.

Primary Analyses

Affect

See Table 2 for regression test statistics and Figure 1 for an illustration of the findings for each study. See Tables 3a, 3b, and 3c for meta-analyses of these results across studies for depressed mood, negative affect, and positive affect, respectively.

Depressed Mood.

In three out of five studies (Studies 2, 3, and 4), we replicated the distraction (vs. rumination) by baseline depression interaction effect on depressed mood reported in the Lyubomirsky, Nolen-Hoeksema, and colleagues’ studies (in support of Hypothesis 1a). Specifically, among those relatively higher in baseline depression (Study 2: BDI ≥ 9.62; Study 3: BDI ≥ 1.66, Study 4: BDI ≥ 7.74), the rumination condition reported higher depressed mood than the distraction condition. Importantly, these cut-offs all fall in the “minimal depression” range (Beck et al., 1996a, 1996b) whereas Lyubomirsky, Nolen-Hoeksema, and colleagues only included those with at least moderate depression in their “depressed group” (those with scores of 20 or above on the current measure). Also in support of Hypothesis 1a, in both the fixed and random effects models, the meta-analysis indicated a significant distraction (vs. rumination) by baseline depression interaction effect on depressed mood across studies.

Testing Hypothesis 2a, in Studies 1 and 4, participants in the gratitude condition reported significantly lower depressed mood than those in the rumination condition, but in Studies 2, 3, and 5, the difference between the two conditions did not reach significance. Overall, in support of Hypothesis 2a, the meta-analysis revealed significant fixed and random effects of gratitude (vs. rumination) such that people in the gratitude condition reported lower depressed mood than those in the rumination condition. We also explored whether these differences were moderated by baseline depression. In Studies 3 and 4, we found a gratitude (vs. rumination) by baseline depression interaction effect on depressed mood such that, among those higher in baseline depression (Study 3: BDI ≥ 4.52; Study 4: BDI ≥ 9.78), the rumination condition reported higher depressed mood than the gratitude condition, but in Studies 1, 2, and 5, the interaction was not significant. In the meta-analysis, we did not find significant fixed or random effects for the gratitude (vs. rumination) by baseline depression interaction on depressed mood. Therefore, we found minimal evidence that the effect of the gratitude (vs. rumination) condition on depressed mood is moderated by baseline depression. The overall difference between the distraction and rumination conditions on depressed mood was also significant in both the fixed and random effects meta-analytic models, but, as noted, was moderated by baseline depression.

Lastly, exploratory analyses did not reveal significant differences between the gratitude and distraction conditions, nor a gratitude (vs. distraction) by baseline depression interaction on depressed mood in any of the individual studies or in the meta-analyses.

Negative Affect.

On negative affect, we found the expected interaction between distraction (vs. rumination) and baseline depression in two studies (Studies 2 and 4) with those in the rumination condition reporting higher negative affect than those in the distraction condition at higher levels of baseline depression (Study 2: BDI ≥ 12.22; Study 4: BDI ≥ 10.01; supporting Hypothesis 1b). In Studies 1 and 5, the rumination condition reported higher negative affect than the distraction condition overall, unmoderated by baseline depression. The meta-analysis supported Hypothesis 1b—both the fixed and random effects models revealed a significant distraction (vs. rumination) by baseline depression interaction effects on depressed mood.

In support of Hypothesis 2b, participants in the gratitude condition reported significantly lower negative affect than those in the rumination condition in all studies except Study 3. The meta-analysis also revealed a significant difference between the two conditions in both the fixed and random effects models. In addition, we explored whether these effects were moderated by baseline depression. We found an interaction between gratitude (vs. rumination) and baseline depression in two studies (Studies 2 and 4) with those in the rumination condition reporting higher negative affect than those in the gratitude condition at higher levels of baseline depression (Study 2: BDI ≥ 9.71; Study 4: BDI ≥ 8.24), but this interaction was not significant in Studies 1, 3, or 5. The fixed effect model in the meta-analysis revealed a significant gratitude (vs. rumination) by baseline depression interaction, but the random effects model did not reach significance. Thus, we found relatively weak evidence that the difference between the gratitude and rumination conditions on negative affect was moderated by baseline depression. The overall difference between the distraction and rumination conditions on negative affect was also significant in both the fixed and random effects meta-analytic models, but, as noted, was moderated by baseline depression in both models.

Lastly, exploratory analyses did not reveal significant differences between the gratitude and distraction conditions, nor a gratitude (vs. distraction) by baseline depression interaction on negative affect in any of the individual studies or in the meta-analyses.

Positive Affect.

In support of Hypothesis 2c, in all five studies the gratitude condition reported higher positive affect than the rumination condition. In support of Hypothesis 3, in four out of five studies (all but Study 3), the gratitude condition reported higher positive affect than the distraction condition. The meta-analysis also supported Hypothesis 2c and 3. Specifically, in support of Hypothesis 2c, both fixed and random effects models revealed that participants in the gratitude condition reported higher positive affect than those in the rumination condition across studies. Similarly, in support of Hypothesis 3, both fixed and random effects models revealed that participants in the gratitude condition reported higher positive affect than those in the distraction condition across studies. The gratitude (vs. rumination) and gratitude (vs. distraction) effects on positive affect were not significantly moderated by baseline depression in any of the individual studies or in the meta-analyses.

Exploratory analyses did not reveal significant differences between the distraction and rumination conditions on positive affect, nor any distraction (vs. rumination) by baseline depression interactions in any of the individual studies or in the meta-analyses.

Analyses of Indirect Effects

Twenty Statements Test

All indirect effects analyses on the Twenty Statements Test are included in Table 4. In both studies (Studies 1 and 3), we found an indirect effect of gratitude (vs. rumination) on coder-rated positivity of thought-action repertoires (higher) and coder-rated negativity of thought-action repertoires (lower) via positive affect (in support of Hypothesis 4b). In Study 1, but not Study 3, we found an indirect effect of gratitude (vs. distraction) on coder-rated positivity (higher) and coder-rated negativity (lower) via positive affect (in partial support of Hypothesis 4b). In Study 1, but not Study 3, we also found an indirect effect of gratitude (vs. rumination) on coder-rated positivity and coder-rated negativity via negative affect such that reductions in negative affect in the gratitude (vs. rumination) condition were associated with higher coder-rated positivity and lower coder-rated negativity (in partial support of Hypothesis 4a). In Study 1, but not Study 3, we also found an indirect effect of distraction (vs. rumination) on coder-rated positivity and coder-rated negativity via negative affect (in partial support of Hypothesis 4a). Lastly, in Study 3, but not Study 1, increases in positive affect in the gratitude (vs. rumination) condition were associated with higher number of responses overall (in partial support of Hypothesis 4b). Thus, supporting Hypothesis 4b, our most robust finding that replicated across both studies was that the gratitude condition promoted higher positive affect than the rumination condition, and that higher positive affect was associated with higher coder-rated positivity and lower coder-rated negativity of thought-action repertoires. That said, also across both studies, the direct effect of gratitude (vs. rumination) and gratitude (vs. distraction) on coder-rated positivity remained significant, demonstrating that positive affect did not explain all of the differences between those conditions in positivity of thought-action repertoires.

Cognitive Bias Questionnaire

All indirect effects analyses on the Cognitive Bias Questionnaire are included in Table 5. In Study 1, we found no indirect effects of condition on depressed/distorted thinking or on nondepressed/nondistorted thinking via positive or negative affect (failing to support Hypotheses 4a and 4b). In Study 5, we found indirect effects of gratitude (vs. rumination) and distraction (vs. rumination) on depressed/distorted thinking and nondepressed/nondistorted thinking via negative, but not positive affect (in support of Hypotheses 4a, but failing to support Hypothesis 4b). Specifically, in Study 5 (but not Study 1), people in both the gratitude and distraction conditions reported lower negative affect than those in the rumination condition, and lower negative affect was related to lower depressed/distorted thinking and higher nondepressed/nondistorted thinking. Importantly, positive affect did not uniquely explain depressed/distorted or nondepressed/nondistorted thinking beyond negative affect in either study.

Table 5.

Indirect Effect of Condition on Cognitive Bias Questionnaire via Positive and Negative Affect (Study 1 and Study 5)

	Gratitude vs. Rumination		Distraction vs. Rumination		Gratitude vs. Distraction
	Effect (SE)	95% CI	Effect (SE)	95% CI	Effect (SE)	95% CI
STUDY 1
Depressed/Distorted Thinking
Positive Affect
a	0.53 (0.13)***	[0.28, 0.78]	−0.22 (0.13) †	[−0.47, 0.04]	0.74 (0.13)***	[0.49, 1.00]
b	0.02 (0.16)	[−0.30, 0.34]	0.02 (0.16)	[−0.30, 0.34]	0.02 (0.16)	[−0.30, 0.34]
Indirect Effect (ab)	0.01 (0.09)	[−0.18, 0.20]	−0.004 (0.04)	[−0.10, 0.08]	0.01 (0.13)	[−0.23, 0.26]
Negative Affect
a	−0.24 (0.10)*	[−0.44, −0.04]	−0.33 (0.10)**	[−0.53, −0.13]	0.09 (0.10)	[−0.11, 0.29]
b	0.08 (0.20)	[−0.32, 0.49]	0.08 (0.20)	[−0.32, 0.49]	0.08 (0.20)	[−0.32, 0.49]
Indirect Effect (ab)	−0.02 (0.05)	[−0.11, 0.07]	−0.03 (0.06)	[−0.16, 0.09)	0.01 (0.03)	[−0.03, 0.08]
Direct Effect (c’)	0.46 (0.36)	[−0.26, 1.18]	−0.03 (0.36)	[−.74, 0.69]	0.49 (0.38)	[−0.25, 1.23]
Nondepressed/Nondistorted Thinking (CBQ)
Positive Affect
a	0.53 (0.13)***	[0.28, 0.78]	−0.22 (0.13)†	[−0.47, 0.04]	0.74 (0.13)***	[0.49, 1.00]
b	0.16 (0.24)	[−0.30, 0.62]	0.16 (0.24)	[−0.30, 0.62]	0.16 (0.24)	[−0.30, 0.62]
Indirect Effect (ab)	0.08 (0.13)	[−0.17, 0.35]	−0.03 (0.06)	[−0.18, 0.08]	0.12 (0.17)	[−0.22, 0.47]
Negative Affect
a	−0.24 (0.10)*	[−0.44, −0.04]	−0.33 (0.10)**	[−0.53, −0.13]	0.09 (0.10)	[−0.11, 0.29]
b	−0.26 (0.30)	[−0.85, 0.33]	−0.26 (0.30)	[−0.85, 0.33]	−0.26 (0.30)	[−0.85, 0.33]
Indirect Effect (ab)	0.06 (0.07)	[−0.05, 0.21]	0.09 (0.08)	[−0.07, 0.27]	−0.02 (0.04)	[−0.12, 0.04]
Direct Effect (c’)	−0.58 (0.53)	[−1.63, 0.47]	0.50 (0.53)	[−0.54, 1.54]	−1.08 (0.55)†	[−2.16, 0.002]
STUDY 5
Depressed/Distorted Thinking
Positive Affect
a	0.56 (0.10)***	[0.38, 0.75]	−0.05 (0.09)	[−0.23, 0.14]	0.61 (0.10)***	[0.42, 0.80]
b	−0.25 (0.15)†	[−0.54, 0.05]	−0.25 (0.15)†	[−0.54, 0.05]	−0.25 (0.15)†	[−0.54, 0.05]
Indirect Effect (ab)	−0.14 (0.09)	[−0.34, 0.02]	0.01 (0.03)	[−0.04, 0.07]	−0.15 (0.10)	[−0.36, 0.03]
Negative Affect
a	−0.27 (0.07)***	[−0.40, −0.14]	−0.21 (0.07)**	[−0.35, −0.08]	−0.06 (0.07)	[−0.19, 0.07]
b	0.69 (0.21)**	[0.28, 1.10]	0.69 (0.21)**	[0.28, 1.10]	0.69 (0.21)**	[0.28, 1.10]
Indirect Effect (ab)	−0.19 (0.09)	[−0.37, −0.04]	−0.15 (0.08)	[−0.32, −0.02]	−0.04 (0.05)	[−0.14, 0.04]
Direct Effect (c’)	−0.11 (0.30)	[−0.69, 0.48]	0.27 (0.29)	[−0.29, 0.84]	−0.38 (0.30)	[−0.97, 0.21]
Nondepressed/Nondistorted Thinking
Positive Affect
a	0.56 (0.10)***	[0.38, 0.75]	−0.05 (0.09)	[−0.23, 0.14]	0.61 (0.10)***	[0.42, 0.80]
b	0.07 (0.24)	[−0.40, 0.53]	0.07 (0.24)	[−0.40, 0.53]	0.07 (0.24)	[−0.40, 0.53]
Indirect Effect (ab)	0.04 (0.13)	[−0.22, 0.32]	−0.003 (0.02)	[−0.06, 0.05]	0.04 (0.15)	[−0.25, 0.34]
Negative Affect
a	−0.27 (0.06)***	[−0.40, −0.14]	−0.21 (0.07)**	[−0.35, −0.08]	−0.06 (0.07)	[−0.19, 0.07]
b	−1.40 (0.34)***	[−2.06, −0.74]	−1.40 (0.34)***	[−2.06, −0.74]	−1.40 (0.34)***	[−2.06, −0.74]
Indirect Effect (ab)	0.38 (0.13)	[0.15, 0.68]	0.30 (0.12)	[0.09, 0.58]	0.08 (0.09)	[−0.09, 0.26]
Direct Effect (c’)	−0.29 (0.48)	[−1.22, 0.65]	−0.62 (0.46)	[−1.53, 0.29]	0.33 (0.48)	[−0.61, 1.27]

Note. See note in Table 4 for coding scheme. For indirect effects analyses, confidence intervals are considered significant if they do not include zero and are designated in bold.

Event Frequency Task

All indirect effects analyses on the Event Frequency Task are included in Table 6. In both studies (Studies 2 and 3), we found an indirect effect of gratitude (vs. rumination) on positive event frequency (higher) via positive affect (supporting Hypothesis 4b). In Study 2, but not Study 3, we also found an indirect effect of gratitude (vs. distraction) on positive event frequency (higher) via positive affect (supporting Hypothesis 4b). That said, Hypothesis 4b was not supported on negative event frequency, as we found no indirect effect via positive affect in either study. In addition, failing to support Hypothesis 4a, we found no indirect effect of condition on event frequency (positive or negative) via negative affect. In sum, Hypothesis 4b was largely supported for positive event frequency—higher positive affect in the gratitude condition was associated with people construing their life events more favorably, but Hypothesis 4a (via negative affect) was not supported.

Table 6.

Indirect Effect of Condition on Event Frequency Task via Positive and Negative Affect (Studies 2 and 3)

	Gratitude vs. Rumination		Distraction vs. Rumination		Gratitude vs. Distraction
	Effect (SE)	95% CI	Effect (SE)	95% CI	Effect (SE)	95% CI
STUDY 2
Negative Events
Positive Affect
a	0.57 (0.15)***	[0.28, 0.86]	−0.06 (0.15)	[−0.35, 0.23]	0.63 (0.15)***	[0.33, 0.92]
b	0.01 (0.03)	[−0.04, 0.07]	0.01 (0.03)	[−0.04, 0.07]	0.01 (0.03)	[−0.04, 0.07]
Indirect Effect (ab)	−0.01 (0.02)	[−0.03, 0.05]	−0.001 (0.01)	[−0.01, 0.01]	0.01 (0.02)	[−0.04, 0.06]
Negative Affect
a	−0.29 (0.11)**	[−0.50, −0.08]	−0.24 (0.11)*	[−0.44, −0.03]	−0.05 (0.11)	[−0.26, 0.16]
b	0.05 (0.04)	[−0.03, 0.14]	0.05 (0.04)	[−0.03, 0.14]	0.05 (0.04)	[−0.03, 0.14]
Indirect Effect (ab)	−0.02 (0.01)	[−0.05, 0.01]	−0.01 (0.01)	[−0.04, 0.01]	−0.003 (0.01)	[−0.02, 0.01]
Direct Effect (c’)	−0.14 (0.07)†	[−0.28, 0.001]	−0.06 (0.07)	[−0.20, 0.08]	−0.08 (0.07)	[−0.22, 0.07]
Positive Events
Positive Affect
a	0.57 (0.15)***	[−0.28, 0.86]	−0.06 (0.15)	[−0.35, 0.23]	0.63 (0.15)***	[0.33, 0.92]
b	0.08 (0.03)**	[−0.02, 0.14]	0.08 (0.03)**	[−0.02, 0.14]	0.08 (0.03)**	[−0.02, 0.14]
Indirect Effect (ab)	0.05 (0.02)	[0.01, 0.09]	−0.005 (0.01)	[−0.03, 0.02]	0.05 (0.02)	[0.01, 0.10]
Negative Affect
a	−0.29 (0.11)**	[−0.50, −0.08]	−0.24 (0.11)*	[−0.44, −0.03]	−0.05 (0.11)	[−0.26, 0.16]
b	0.01 (0.04)	[−0.08, 0.09]	0.01 (0.04)	[−0.08, 0.09]	0.01 (0.04)	[−0.08, 0.09]
Indirect Effect (ab)	−0.002 (0.01)	[−0.03, 0.02]	−0.002 (0.01)	[−0.03, 0.02]	−0.0003 (0.01)	[−0.01, 0.01]
Direct Effect (c’)	0.08 (0.07)	[−0.06, 0.23]	0.12 (0.07)	[−0.02, 0.26]	−0.03 (0.07)	[−0.18, 0.11]
STUDY 3
Negative Events (EFT)
Positive Affect
a	0.21 (0.08)**	[0.05, 0.36]	0.10 (0.08)	[−0.06, 0.25]	0.11 (0.08)	[−0.04, 0.26]
b	0.04 (0.03)	[−0.01, 0.09]	0.04 (0.03)	[−0.01, 0.09]	0.04 (0.03)	[−0.01, 0.09]
Indirect Effect (ab)	0.007 (0.004)	[−0.002, 0.01]	0.004 (0.004)	[−0.002, 0.01]	0.004 (0.005)	[−0.002, 0.02]
Negative Affect
a	−0.09 (0.06)	[−0.22, 0.03]	−0.10 (0.06)	[−0.22, 0.02]	0.005 (0.06)	[−0.11, 0.13]
b	0.18 (0.03)***	[0.12, 0.25]	0.18 (0.03)***	[0.12, 0.25]	0.18 (0.03)***	[0.12, 0.25]
Indirect Effect (ab)	−0.02 (0.01)	[−0.04, 0.002]	−0.02 (0.01)	[−0.05, 0.01]	0.001 (0.01)	[−0.02, 0.02]
Direct Effect (c’)	−0.09 (0.05)†	[−0.19, 0.002]	−0.08 (0.05)†	[−0.18, 0.01]	−0.01 (0.05)	[−0.11, 0.08]
Positive Events (EFT)
Positive Affect
a	0.21 (0.08)**	[0.05, 0.36]	0.10 (0.08)	[−0.06, 0.25]	0.11 (0.08)	[−0.04, 0.26]
b	0.17 (0.03)***	[0.12, 0.22]	0.17 (0.03)***	[0.12, 0.22]	0.17 (0.03)***	[0.12, 0.22]
Indirect Effect (ab)	0.04 (0.01)	[0.01, 0.06]	0.02 (0.01)	[−0.01, 0.04]	0.02 (0.01)	[−0.01, 0.05]
Negative Affect
a	−0.09 (0.06)	[−0.22, 0.03]	−0.10 (0.06)	[−0.22, 0.02]	0.01 (0.06)	[−0.11, 0.13]
b	0.02 (0.03)	[−0.04, 0.09]	0.02 (0.03)	[−0.04, 0.09]	0.02 (0.03)	[−0.04, 0.09]
Indirect Effect (ab)	−0.002 (0.004)	[−0.01, 0.005]	−0.002 (0.004)	[−0.01, 0.01]	0.0001 (0.003)	[−0.01, 0.01]
Direct Effect (c’)	0.02 (0.05)	[−0.08, 0.12]	−0.04 (0.05)	[−0.14, 0.07]	0.05 (0.05)	[−0.05, 0.15]

Note. See note in Table 4 for coding scheme. For indirect effects analyses, confidence intervals are considered significant if they do not include zero and are designated in bold.

Means Ends Problem Solving Task

We found no indirect effects of condition on variables from the Means Ends Problem Solving Task via positive or negative affect (see Table 7, only administered in Study 2).

Table 7.

Indirect Effect of Condition on Means Ends Problem Solving Task via Positive and Negative Affect (Study 2)

	Gratitude vs. Rumination		Distraction vs. Rumination		Gratitude vs. Distraction
	Effect (SE)	95% CI	Effect (SE)	95% CI	Effect (SE)	95% CI
Overall Effectiveness on Story 1
Positive Affect
a	0.57 (0.15)***	[0.28, 0.86]	−0.06 (0.15)	[−0.35, 0.23]	0.63 (0.15)***	[0.33, 0.92]
b	−0.06 (0.10)	[−0.13, 0.25]	−0.06 (0.10)	[−0.13, 0.25]	−0.06 (0.10)	[−0.13, 0.25]
Indirect Effect (ab)	0.03 (0.06)	[−0.09, 0.16]	−0.003 (0.02)	[−0.04, 0.03]	0.04 (0.07)	[−0.10, 0.16]
Negative Affect
a	−0.29 (0.11)**	[−0.50, −0.08]	−0.24 (0.11)*	[−0.44, −0.03]	−0.05 (0.11)	[−0.26, 0.16]
b	0.18 (0.14)	[−0.09, 0.44]	0.18 (0.14)	[−0.09, 0.44]	0.18 (0.14)	[−0.09, 0.44]
Indirect Effect (ab)	−0.05 (0.04)	[−0.15, 0.03]	−0.04 (0.04)	[−0.13, 0.02]	−0.01 (0.02)	[−0.06, 0.03]
Direct Effect (c’)	−0.44 (0.23)†	[−0.90, 0.02]	−0.24 (0.23)	[−0.69, 0.20]	−0.19 (0.24)	[−0.66, 0.27]
Overall Effectiveness on Story 2
Positive Affect
a	0.57 (0.15)***	[0.28, 0.86]	−0.06 (0.15)	[−0.35, 0.23]	0.63 (0.15)***	[0.33, 0.92]
b	0.18 (0.10)†	[−0.02, 0.39]	0.18 (0.10)†	[−0.02, 0.39]	0.18 (0.10)†	[−0.02, 0.39]
Indirect Effect (ab)	0.10 (0.06)	[−0.01, 0.24]	−0.01 (0.03)	[−0.08, 0.05]	0.12 (0.07)	[−0.01, 0.27]
Negative Affect
a	−0.29 (0.11)**	[−0.50, −0.08]	−0.24 (0.11)*	[−0.44, −0.03]	−0.05 (0.11)	[−0.26, 0.16]
b	−0.04 (0.14)	[−0.32, 0.25]	−0.04 (0.14)	[−0.32, 0.25]	−0.04 (0.14)	[−0.32, 0.25]
Indirect Effect (ab)	0.01 (0.04)	[−0.08, 0.11]	0.01 (0.04)	[−0.07, 0.09]	0.002 (0.02)	[−0.03, 0.04]
Direct Effect (c’)	−0.19 (0.24)	[−0.68, 0.30]	0.10 (0.24)	[−0.38, 0.57]	−0.29 (0.25)	[−0.78, 0.20]
Percentage Effectiveness Across Four Stories
Positive Affect
a	0.53 (0.15)***	[0.24, 0.83]	−0.10 (0.15)	[−0.40, 0.19]	0.63 (0.15)***	[0.34, 0.93]
b	−0.01 (0.01)	[−0.03, 0.01]	−0.01 (0.01)	[−0.03, 0.01]	−0.01 (0.01)	[−0.03, 0.01]
Indirect Effect (ab)	−0.004 (0.01)	[−0.02, 0.01]	0.001 (0.002)	[−0.003, 0.01]	−0.004 (0.01)	[−0.02, 0.01]
Negative Affect
a	−0.27 (0.11)*	[−0.48, −0.06]	−0.21 (0.11)*	[−0.42, −0.004]	−0.05 (0.11)	[−0.26, 0.16]
b	−0.01 (0.02)	[−0.03, 0.02]	−0.01 (0.02)	[−0.03, 0.02]	−0.01 (0.02)	[−0.03, 0.02]
Indirect Effect (ab)	0.001 (0.004)	[−0.01, 0.01]	0.001 (0.003)	[−0.01, 0.01]	0.0003 (0.002)	[−0.003, 0.004]
Direct Effect (c’)	−0.04 (0.03)†	[−0.09, 0.01]	−0.03 (0.03)	[−0.08, 0.02]	−0.01 (0.03)	[−0.07, 0.04]

Note. See note in Table 4 for coding scheme. For indirect effects analyses, confidence intervals are considered significant if they do not include zero and are designated in bold.

Judgment of Pleasant Activities

All indirect effects analyses on Judgment of Pleasant Activities are included in Table 8. In support of Hypothesis 4b, but not 4a, in Study 4, we found indirect effects of gratitude (vs. rumination) and gratitude (vs. distraction) on how enjoyable people rated a list of pleasant activities and how likely they were to engage in them via positive affect, but not negative affect (see Table 8, only administered in Study 4). Thus, supporting our prediction, higher positive affect in the gratitude condition was associated with people rating potential activities more favorably.

Summary of Analyses in Supplemental Material

In Supplemental Material, we included multiple regression analyses with condition, baseline depression, and their interaction predicting our thought pattern outcome variables (without the indirect effects via affect). Across both studies (Study 1 and 3), participants in the gratitude condition had higher coder-rated positivity of thought-action repertoires than those in the rumination and distraction conditions, and participants in the distraction condition had lower coder-rated negativity of thought-action repertoires than those in the rumination condition. No other results were consistent across two studies, so we do not summarize them here.

In addition to the coder-rated positivity and negativity of thought-action repertoires included in the main manuscript, we also explored the effects of condition and baseline depression on mutually exclusive categories representing what the participant said they wanted to do at that moment (e.g., sleep/rest, schoolwork/work, exercise/sport, outdoors/nature, express gratitude, be social). Across both studies (Study 1 and 3), people in the gratitude condition were significantly more likely to mention wanting to be social and express gratitude than those in the rumination and distraction conditions. Also across both studies, people in the gratitude condition were more likely than those in the distraction condition to mention wanting to self-improve in some way (e.g., be a better person).

We also tested the effect of condition and baseline depression on grateful feelings (a composite of thankful, grateful, and appreciative) and indebtedness (a single item). In all five studies, participants in the gratitude condition reported higher grateful feelings than those in the distraction or rumination conditions. In addition, we found that participants in the gratitude condition reported similar indebtedness to those in the rumination condition (in four out of five studies), but higher indebtedness than those in the distraction condition (in four out of five studies).

To more closely resemble Lyubomirsky, Nolen-Hoeksema, and colleagues’ studies, we also explored baseline depression as dichotomous (i.e., depressed or nondepressed) and applied the same contrast analyses that they used in their studies (e.g., −3: depressed/rumination, +1: depressed/distraction, +1: nondepressed/rumination, +1: nondepressed/distraction). Across studies, we replicated their findings on depressed mood. We also often replicated their findings on the thought pattern outcomes (in 8 out of 13 tests), but a closer examination of the means revealed that people with depression who distracted themselves sometimes had either equal or more negative thought patterns than those who ruminated and the significant contrasts were driven by nondepressed participants reporting healthier thinking patterns.

In Study 3, we included a second time point, one week after the first, and explored whether our condition effects on affect and event frequency lasted. Participants in the distraction condition reported lower negative affect than those in the rumination condition even one week later and this effect was still stronger among those higher in baseline depression. On depressed mood, the distraction (vs. rumination) by baseline depression interaction was also still significant. We found no significant effects on event frequency one week later. In addition, we found no condition effects on life satisfaction or brooding and reflection (a rumination measure) one week later.

Lastly, in all studies, we explored whether gender moderated our effects on affect or the thought pattern outcomes. We found some interactions between condition, baseline depression, and gender, but they were inconsistent across studies and outcomes.

General Discussion

Overview of Results

The effects of condition on positive and negative affect and depressed mood were consistent across studies and in line with our predictions. We will focus on the meta-analytic results here to provide a global summary of the affect findings. As predicted, the meta-analyses revealed a significant distraction (vs. rumination) by baseline depression interaction on depressed mood and negative affect such that participants in the rumination condition reported higher depressed mood and negative affect than those in the distraction condition, especially when they had relatively higher baseline depression (supporting Hypothesis 1a [on depressed mood] and 1b [on negative affect]). This interaction on depressed mood replicated those in the Lyubomirsky, Nolen-Hoeksema and colleagues’ studies.

In support of Hypothesis 2a and 2b, both the fixed and random effects meta-analytic models revealed that participants in the gratitude condition reported lower depressed mood and negative affect, respectively, than those in the rumination condition. In addition, the meta-analysis did not indicate a significant gratitude (vs. rumination) by baseline depression interaction on depressed mood, and only the fixed effect model indicated this interaction on negative affect. Thus, we found relatively weak evidence that the difference between the gratitude and rumination condition on depressed mood and negative affect was moderated by baseline depression. Importantly, the overall differences between the distraction and rumination conditions on both depressed mood and negative affect were also significant, but, as noted, those effects were moderated by baseline depression.

In support of Hypothesis 2c, participants in the gratitude condition reported higher positive affect than those in the rumination condition and, in support of Hypothesis 3, participants in the gratitude condition also reported higher positive affect than those in the distraction condition; these effects were consistent across levels of baseline depression. In sum, participants in the distraction condition reported lower depressed mood and negative affect than those in the rumination condition, and these effects were stronger among those higher in baseline depression. In addition, participants in the gratitude condition reported lower depressed mood and negative affect, and higher positive affect than those in the rumination condition across levels of baseline depression, as well as higher positive affect than those in the distraction condition. Thus, not only did gratitude promote lower negative affect than the rumination condition, it also promoted higher positive affect, which we found to uniquely predict our thought pattern outcomes in our indirect effects analyses.

Specifically, supporting Hypothesis 4b, we found an indirect effect of the gratitude (vs. rumination) condition on six (out of twelve) of our thinking pattern outcomes via positive affect (i.e., number of thought-action repertoires reported [Study 3], coder-rated positivity of thought-action repertoires [Studies 1 and 3], coder-rated negativity of thought-action repertoires [Studies 1 and 3], positive event frequency [Studies 2 and 3], enjoyability of pleasant activities [Study 4], and likelihood of engaging in pleasant activities [Study 4]), indicating that changes in positive affect uniquely explained variance in these thought patterns over and above changes in negative affect. Also supporting Hypothesis 4b, we found an indirect effect of gratitude (vs. distraction) on five of these outcomes via positive affect (i.e., coder-rated positivity of thought-action repertoires [Study 1], coder-rated negativity of thought-action repertoires [Study 1], positive event frequency [Study 2], enjoyability of pleasant activities [Study 4], and likelihood of engaging in pleasant activities [Study 4]). Notably, the indirect effects of the gratitude (vs. distraction) condition were not replicated in a second study, whereas the indirect effects of the gratitude (vs. rumination) condition were replicated when possible (the Judgment of Pleasant Activities measure only appeared in one study). Thus, gratitude (vs. rumination) consistently had an indirect effect on our thought pattern outcomes via positive affect and seemed to relate to those outcomes that were more positive in nature (e.g., an open-ended thought-action repertoire task, reflection on positive event frequency, and judgment of pleasant activities).

Supporting Hypothesis 4a, we found indirect effects of the gratitude (vs. rumination) condition and the distraction (vs. rumination) condition on four of our thought pattern outcomes via negative affect (i.e., coder-rated positivity of thought-action repertoires [Study 1], coder-rated negativity of thought-action repertoires [Study 1], depressed/distorted thinking [Study 5], and nondepressed/nondistorted thinking [Study 5]). That said, none of these indirect effects via negative affect were replicated in a second study. We also found no indirect effects of condition on negative event frequency or the three outcomes in our interpersonal problem solving measure (Means-Ends Problem Solving Task) via either positive or negative affect.

In sum, gratitude promoted higher positive affect than distraction and rumination conditions and, like distraction, also promoted lower negative affect and depressed mood than the rumination condition. Importantly, gratitude-induced changes in positive affect explained differences in our thought pattern outcomes more frequently than gratitude- or distraction-induced changes in negative affect. This evidence points to gratitude being a preferable emotion-regulation strategy to both distraction and rumination, although distraction also demonstrates clear benefits over rumination.

Comparison to Original Rumination Studies

What constitutes a successful replication is often debatable (e.g., Wilson et al., 2020; Open Science Collaboration, 2015) and, in the current studies, we have an added layer of complication in evaluating our replication of the Lyubomirsky, Nolen-Hoeksema, and colleagues’ studies because we treated baseline depression as continuous in the main manuscript and they treated it as dichotomous in their studies (moderately depressed vs. nondepressed, with mildly depressed excluded; but see our dichotomous analyses in Supplemental Material). Despite this difference, our studies replicated their findings on depressed mood—the only affect measure included in Lyubomirsky, Nolen-Hoeksema, and colleagues’ studies. Specifically, our meta-analysis revealed a significant interaction in the expected direction—the participants in the rumination condition reported higher depressed mood than those in the distraction condition, especially among people with relatively higher baseline depression.

Our replication attempts on the thought pattern measures from Lyubomirsky, Nolen-Hoeksema, and colleagues’ studies were less successful (Cognitive Bias Questionnaire, Event Frequency Task, Means-Ends Problem Solving Task, Judgment of Pleasant Activities; see Supplemental Material for specific results [without indirect effect analyses]). In our regression analyses, we found no significant effects of distraction (vs. rumination) or distraction (vs. rumination) by baseline depression interactions on the Cognitive Bias Questionnaire, Means-Ends Problem Solving Task, Judgment of Pleasant Activities, or positive event frequency. We found a distraction (vs. rumination) by baseline depression interaction on negative event frequency in Study 2 that was opposite of the predicted direction—people with relatively lower baseline depression scores reported fewer negative events in the distraction condition than in the rumination condition, but the two conditions were not significantly different at higher levels of baseline depression (this effect was not replicated in Study 3).

Thus, although our results support replication of the depressed mood findings from past work, they do not support replication on these thought pattern outcomes. Importantly, the non-replication on these outcomes could be due to us including people with mild depression in the “depression” group whereas Lyubomirsky, Nolen-Hoeksema, and colleagues only included people with moderate depression or higher (see also results with dichotomized depression in Supplemental Material).

Unique Patterns for Positive and Negative Affect

A major strength of this project is its focus on positive affect as separable from negative affect and as an important contributor to well-being. As psychologist Walter Dearborn once said, “If you want to understand something, try to change it” (in Bronfenbrenner, 1981, p. 37). We add to correlational studies demonstrating the independence of positive affect (e.g., Diener & Emmons, 1985) by showing that our gratitude condition dampened negative affect and improved positive affect, whereas a distraction condition dampened negative affect and either dampened or had no effect on positive affect (see Table 1), demonstrating that positive and negative affect do not always change inversely. Because effect sizes indicated that gratitude was typically as effective as distraction in dampening negative affect compared to the rumination condition, but more effective in enhancing positive affect, these two conditions (gratitude and distraction) are interesting in that they can be used in future research to explore the unique effects of enhancing positive affect above and beyond decreasing negative affect.

Our studies also demonstrate the unique contribution of positive affect—beyond negative affect—to positive construal of people’s lives and their thought-action repertoires (i.e., their thoughts about what they would like to do next). Specifically, across two studies, gratitude-induced changes in positive affect related to more positive construal of the events in one’s life, as well as higher positivity and lower negativity in one’s thought-action repertoires. Analyses of the specific content of thought-action repertoires indicated that, across two studies, people in the gratitude condition were more likely than those in the rumination or distraction conditions to spontaneously mention wanting to be social or express gratitude to someone, both activities that could promote stronger relationships, feeding back into well-being (see Supplemental Material).

Thus, we provide support for the idea that increasing positive affect should be a direct goal of treatments for depression, in addition to reducing negative affect (Layous et al., 2011). For example, Cognitive Behavioral Therapy (CBT; a manualized therapeutic program focused on reducing deficits like negative thoughts, emotions, and behaviors) has been shown in many studies to be a highly effective treatment for mild to moderate depression (Cuipers et al., 2013), but we think an added focus on increasing positive affect may bolster its effects and promote a fuller scope of well-being. This hypothesis is supported in part by a randomized controlled trial comparing CBT to Positive Psychotherapy (PPT; a manualized therapeutic program focused on increasing positive resources like positive affect), in which researchers found that PPT was more effective in reducing depressive symptoms than CBT (Furchtlehner et al., 2020). That said, future research would do well to employ a three-arm comparison of CBT+PPT, CBT, and PPT to see if a combined approach is more effective than either approach alone. For now, consistent with a personalized treatment approach, PPT may be a good adjunct treatment for individuals who are not responding to other psychotherapeutic treatments.

Lastly, Layous and colleagues (2014) speculated that positive activities and the positive affect they give rise to may be protective—mitigating risk factors before they result in psychopathology. Importantly, intentionally thinking gratefully is an activity anyone can do at any time, without cost—it does not have to be practiced in the context of therapy which people may not have the resources to enter. This focus on gratitude (or other positive activities like performing kind acts or thinking optimistically) may elicit momentary positive affect that could help people take action to build important personal resources in their lives (e.g., Fredrickson et al., 2008). These resources (e.g., mindfulness, positive relationships) may help people cope with life’s inevitable challenges, making it less likely that stressors set off a downward spiral toward depression. Thus, not only do we think that positive affect can be effective in mitigating existing depression, we also think it can promote resilience in the face of challenges that might otherwise have triggered depression.

A New Gratitude Manipulation

In an attempt to mirror Lyubomirsky, Nolen-Hoeksema, and colleagues’ rumination and distraction conditions, we came up with a new gratitude manipulation in which people contemplate a list of 45 statements for which they might be grateful (e.g., “Think about the people in your life who value you”). This new gratitude intervention successfully boosted positive affect when compared to both the rumination and distraction conditions. Importantly, these condition effects were not moderated by baseline depression, demonstrating the efficacy of the intervention in promoting positive affect across levels of depression (see also Supplemental Material for results on grateful feelings specifically). The gratitude condition also reported lower negative affect than the rumination condition, demonstrating its potential efficacy in reducing negative affect.

Past research has found that gratitude manipulations do not consistently reduce negative affect when compared to neutral conditions (Dickens, 2017). In the current studies, due to the distraction condition dampening negative affect and the rumination condition increasing it overall, we do not have a true neutral comparison condition. That said, we speculate that this gratitude condition may not elicit as many mixed feelings (e.g., positive affect, but also indebtedness) as gratitude conditions that ask participants to focus on one specific target (e.g., gratitude letter). A future study is needed to directly compare the current gratitude condition to past gratitude manipulations (e.g., gratitude letters, gratitude experiences, counting blessings, three good things) to explore whether it evokes a less mixed emotional experience.

One potential reason this gratitude condition was so effective in increasing positive affect and decreasing negative affect might be because it had 45 statements, any of which could resonate with the participant, allowing for a greater chance of person-activity fit than activities focused on just one target of gratitude (Lyubomirsky & Layous, 2013). Also, reflecting on the 45-statement list may have been less cognitively burdensome and more enjoyable than coming up with one’s own target of gratitude and writing about them (cf. Lyubomirsky et al., 2006). That said, past research has found long-term effects from a single instance of a gratitude letter or counting blessings (“three good things”) activity (Seligman et al., 2005), whereas the effects of the current gratitude manipulation did not last beyond one week (see Supplemental Material for analyses on Time 2 in Study 3). As mentioned, future research would do well to directly compare the current gratitude manipulation to past gratitude manipulations if practiced weekly over time.

Limitations and Future Directions

Although our set of studies had strengths like consistent replication of our affect findings across our five studies and a diverse set of samples, we also had limitations. For example, all of our measures were self-reported, which is appropriate for measuring affect, but self-reported measures of interpersonal problem solving (Means-Ends Problem Solving Task) or thought-action repertoires (The Twenty Statements Test) likely do not fully capture how people would address interpersonal problems in real life or what they would actually do following the experimental manipulation. Future research could employ creative methods to assess behavior following our manipulation. For instance, couples could come into the laboratory, be randomly assigned to a certain condition, and then discuss a recurring problem in their relationship. We would expect that those in the gratitude condition would be more likely to inject humor and lightheartedness into the conversation and less likely to be critical and defensive, undermining a resolution. Alternatively, following the experimental manipulation, researchers could see who is most likely to sign up for an exercise class or volunteer opportunity. Ideally, we want to tie experimentally-induced changes in affect to actual behavior that may feed back into well-being, and so far, we have only captured thoughts, not behaviors. That said, given that negative thinking patterns are a key aspect of depression, capturing thought patterns is also important even if not the gold standard (e.g., Baumeister et al., 2007). In addition, measures like the Twenty Statements Test that require open-ended responses that are then coded by independent raters can be considered performative and not subject to as many biases as self-report scales.

Similarly, although we purport to explore the effects of the gratitude, rumination, and distraction conditions across levels of baseline depression, our measure of baseline depression was also self-reported and our participants were not evaluated by a licensed professional to verify their scores. Although our diverse student samples had depression scores comparable to other similar samples (e.g., Whisman & Richardson, 2015), our online participants from Study 3 had much lower scores than a comparable community sample (Segal et al., 2008). Indeed, over 40% of the sample in Study 3 indicated no depressive symptoms at all (i.e., a score of zero), which seems unlikely. Perhaps some of the weak findings from Study 3 could be attributed to participants not answering questions honestly or carefully. Alternatively, Study 3 participants’ baseline depression scores could be truly lower than the other samples and the weak findings could be because some of our effects were stronger at higher levels of depression. Future research could employ our conditions on a sample of people with clinical diagnoses to verify our findings.

In addition to the relatively low baseline depression scores in Study 3, we also had concerns over participant attrition and attention in that study (see Supplemental Material). Although we still found some of our expected condition effects on affect at post-test, they were not as strong as in the other studies and most did not last one week later. Unfortunately, this was the only study in which we included a second time point to explore the duration of our condition effects. Importantly, although Layous and colleagues (2014) speculated that positive activities like expressing gratitude could mitigate rumination and interrupt a downward spiral into depression, we found no effects of gratitude on rumination (brooding) one week later (see results for Study 3, Time 2 in Supplemental Material). Given the issues with the sample and the weaker results at Time 1, this was not the ideal study in which to explore the longevity of the effects. Future research would do well to explore the effects of these conditions after one or more weeks, or the effects of these conditions administered weekly over time (e.g., for six weeks similar to other gratitude interventions).

Finally, although our indirect effects analyses test a process model by which condition influences affect which then predicts differences in thought patterns, we cannot be sure of this causal sequence because we measured all constructs at the same time point. That said, given that we largely did not find condition effects directly on the thought pattern outcomes, the sequence from condition to thought patterns to affect is not as supported as the one from condition to affect to thought patterns. Future researchers should examine the constructs studied here in a longitudinal framework to disentangle these possibilities and identify how these processes unfold over time.

Conclusions

Across five studies, we found remarkably similar results—the gratitude condition mitigated negative affect when compared to the rumination condition and also promoted higher positive affect than the rumination and distraction conditions. Importantly, these differences in affect were related to differences in thought patterns—people in the gratitude (vs. rumination) had higher positive affect, which was consistently related to higher positivity (and lower negativity) in thought-action repertoires and more positive construal of the events in people’s lives. Years ago, Lyubomirsky, Nolen-Hoeksema, and colleagues demonstrated the self-perpetuating cycle of people with depression who ruminated. We replicated their results on depressed mood, but largely did not replicate their findings on their other outcomes (e.g., depressed/distorted thinking) which they posited fed back into depression in a self-perpetuating cycle. Nevertheless, we found a different self-perpetuating cycle—a positive upward spiral. The gratitude condition promoted higher positive affect, which helped people view their lives more positively and express a desire to engage in positive behaviors that could feed back into better mental health.