Introduction

Emotion regulation (ER)—the manners by which individuals influence their emotional states—is appreciated as a dynamic process with sequential components. Gross’s1 extended process model, for instance, discretizes ER into identification, strategy selection, and implementation stages (see also2. As explicated by Sheppes et al.3, “at the heart of each stage is a central emotion-regulation-related decision that needs to be made” (p. 383), and for this reason, a significant portion of ER research has been devoted to understanding how people select among pools of regulatory options (strategies, tactics, etc.) (e.g.4). However, beyond selection mechanisms, also mentioned among these frameworks are notions of “instantiation.” For example, the strategy selection stage models strategy availability as whether a strategy “can be instantiated” in that particular situation or not3 (p. 389), whereas the implementation stage then involves choosing among these instantiated options to be actually applied5. This process has also been illustrated in Gross6 as a decision tree delineating the wide response space of possible regulatory ideas.

Critically, how these response spaces are generated within individuals has not yet been fully investigated in practice. In other words, it is currently unclear what factors influence the raw generation of ER ideas (henceforth referred to as “techniques,” from the taxonomy of Powers and LaBar7. This generative process is important to measure as a unique stage during the ER process because it theoretically follows strategy selection but precedes the actual application of the technique (see8,9 for a similar model specific to reappraisal generation), and thus isolates a relatively underappreciated node of ER that can be targeted in clinical applications. In examining ER generation, this study sought to answer if individuals’ ER profiles could be indexed by the fluency and diversity with which they generate ER techniques, and if these abilities relate to other modes of idea generation as well as depressive symptomology.

Conceptualizing ER as involving a unique generative component is important for several reasons. First, generated techniques capture the situation specificity that has become a growing target for ER research2,10. That is, this generative process more naturalistically reflects the nature by which individuals actually regulate: not through strategy instruction or self-reporting habitual ER tendencies, but rather by personally ideating specific ways to respond to a particular stimulus (e.g., reappraising that “at least I will have more time to exercise” following a job termination). Notably, this ideation process extends across strategies considered both adaptive (e.g., reappraisal) and maladaptive (e.g., rumination). Some paradigms already incorporate elements of this context-specific ideation (e.g., “…come up with possible reasons why someone might give you a certain offer”11 (p. 817), yet these studies often examine only specific strategies to clarify their putative adaptiveness, and do not directly assess the generated techniques themselves.

Secondly, and relatedly, researchers have increasingly called for work that examines “the heterogeneity of ways in which people engage” within specific strategies, and the spontaneity of these regulations10, (p. 160). Free-response paradigms are well-suited to capture both, hence Ladis et al.12(p. 256) suggested that ER researchers leverage such paradigms to obtain qualitative data of ER specificity. Yet, such approaches have not been widely implemented except in specific assessments for the strategy of reappraisal13,14,15,16. These studies also prompted individuals to generate reappraisals to examine qualitative metrics of interest (e.g., reappraisal inventiveness), whereas here we were more interested in what strategies are readily available to individuals, particularly when they are not instructed to use specific strategies. We suspected that a qualitative approach to this question would be feasible given that Waugh et al.9 speculated that similar patterns of generation to reappraisal are involved in other strategies. Indeed, Davis et al.17 demonstrated the feasibility of capturing and coding free-response ER responses according to broader strategies in a child sample. To our knowledge, no study to date has yet implemented a cross-strategy free-response paradigm in adults and measured how such generation patterns are sensitive to individual differences in creative abilities and emotional well-being.

Third, the process of generating techniques itself is important, as it provides tangible practice with the broader strategies under which the techniques belong. This direct experience can then inform other components of ER behavior, including strategy selection18 and effectiveness19. For instance, Sheppes et al.4 reported that participants were more likely to choose reappraisal if the experimenters manipulated its generation process by offering templates of reappraisals (e.g., reframing losses as improvement potential). Thus, individuals with varying generative capacities across strategies may show downstream changes in selection, implementation, and even effectiveness of these strategies.

By measuring ER generation, we are also afforded insight into a component of ER that may characterize affective disorders. Sheppes et al.4 proposed that psychopathology may inhibit generation due to the presence of more rigid thinking styles. This factor may be particularly salient in depression, as a meta-analysis revealed impediments in ideation abilities that seem to worsen with increasing depressive symptoms20. Do these faulty generative processes, however, apply to ER generation in particular? Liu and Thompson21 partly answered this question in their meta-analysis of spontaneous, or non-instructed, ER. They found that, of six common strategies, four had never been inspected with regard to spontaneous generation, and only one (reappraisal) showed similar rates of spontaneous use between depressed and never-depressed individuals. Further, the included reappraisal studies operationalized “spontaneous use” as the degree to which participants retrospectively estimated reappraising during a negative induction22,23,24, rather than a qualitative analysis of the unfolding reappraisal attempts themselves. Thus, research is still needed to elucidate the modulating role of depression on idea generation in ER across strategies.

Perhaps due to practical challenges12, the ER field does not commonly utilize free-response data in its paradigms, which is potentially attributable to the lack of paradigms by which to frame and analyze such rich data. To address this gap, we offer that the domain of creativity is well-equipped to measure open-ended ideation across categories (i.e., strategies). In particular, divergent thinking (DT) is the most common measure of creative aptitude, involving the diverse production of ideas surrounding some prompt25. DT tasks typically yield two variables: fluency (the number of ideas generated) and some measure of novelty. While researchers often use originality (statistical infrequency) to quantify novelty, also common (and relevant to ER) is flexibility, or the “variety of conceptual categories”25 (p. 67). Using these principles, we propose that ER generation across strategies is a DT process: First, ER involves open-ended ideation surrounding a stimulus. Second, generated techniques can be counted (fluency). Third, generated techniques can be classified according to broader organizing categories, or strategies7. While this categorical measurement maps onto the DT notion of flexibility, we refer to this term instead as diversity to avoid confusion with ER frameworks that use “flexibility” with a different meaning (e.g., flexible application of ER strategies given contextual demands).

We subsequently developed a DT task—ERGen—that indexes a novel ER profile by measuring technique generation and categorization according to broader-level strategies using both human raters and a generative pre-trained transformer (GPT) model26. We hypothesized in our preregistration that enhanced creative-thinking ability would associate with greater strategy diversity and fluent technique generations, whereas higher depression symptom severity would reveal the inverse (lower strategy diversity and fewer generated techniques). In addition, because depression can be comprised of both dysphoric (mood) and anhedonic (reward/motivation) components, we also sought to determine if including individual levels of mood and motivation in our models would yield any additional insights on the specific aspect of depression at play during generation.

Further, based on Matheson and Anisman27, we predicted that greater depression severity would lead to greater use of specific strategies like emotional expression, emotional containment, self- and other-blame, and rumination, and lower use of problem-solving, active distraction, and social support seeking. Because of the lack of work comparing ER to creativity, we did not provide any strategy-specific predictions with regard to creative-thinking ability. Lastly, because some have speculated that “trait and spontaneous ER do not frequently generate the same results”21(p. 191), we also compared generation results to self-reported ER tendencies. We anticipated there to be a convergence, but moderated by variables like depression severity such that with higher depression, generation more strongly mimics self-report tendencies due to a lower willingness to experiment with less commonly utilized strategies. If successful, this work will ultimately advance the experimental practice and theoretical modeling of ER by emphasizing idea-generation processes and thus localizing potential new clinical targets for interventions.

Methods

Data availability

The study was approved by the Duke University Campus Institutional Review Board and was performed in accordance with the relevant guidelines and regulations (2025-0009). The study’s design and analysis plan were preregistered. The preregistration, analysis code, data, and materials can be found at https://osf.io/adyh3.

Participants

To establish our sample size, we conducted a power analysis for our most complex preregistered statistical model containing six predictors of interest (see Data analysis) using G*Power28. In order to achieve power of 80%, the power analysis determined a sample size of 143 was needed to detect a significant R2 deviation from zero with a small-medium effect size of 0.1 for a linear multiple regression. Note that this regression model represents a simplified version of our mixed-effects hurdle model, given challenges with estimating sample sizes for such models and lack of standardized methodology. We recruited an online sample of 160 adult individuals on Prolific to account for exclusions due to poor or incomplete data. Inclusion criteria required that participants reside in the United States, speak English as their first language, have an approval rating of at least 90% on Prolific, and have at least 50 studies previously completed on Prolific. Additionally, to avoid potential differences in typing speed across different devices, participants could only enroll in the study if they were completing the task on a computer.

In accordance with our preregistration, we excluded data from five participants who did not provide any answer to four or more of the 20 scenarios in our ER generation task. Five additional participants misunderstood the task instructions and listed only their emotional states (e.g., “Disappointed.”) for four or more of the 20 scenarios; the data from these participants were excluded as well for not reporting ER techniques. We also noticed strikingly similar responses across seven participants, who often wrote in a second-person perspective indicative of a large-language model (LLM) such as ChatGPT (e.g., “Here are some ideas to manage negative emotions in this situation: Acknowledge your feelings of embarrassment; it’s a natural response. Remind yourself that everyone makes mistakes when learning something new…”). Although we did not originally specify this exclusion criteria, the data from these participants were excluded for suspected use of an outside program to generate techniques (e.g., ChatGPT) [We also note that these seven participants showed significantly different patterns of mouse clicks from the final sample, presumably due to copy and pasting responses (see Figure S1). We recommend future studies to eliminate the ability to paste text into free-response boxes through JavaScript code within the Qualtrics questionnaire.]. These exclusions resulted in a final sample size of 143 participants ranging in age from 20 to 81 years old (M = 40.58, SD = 14.07; 81 females, one non-binary). Participants constituted the following racial/ethnic makeup: 78 White or European origin, 39 Black or African American, six East or Southeast Asian, five Hispanic/Latine, one American Indian or Native Alaskan, one South Asian, and 13 bi- or multi-racial participants. Participants were debriefed following completion of the study and were compensated $15 for their time, with a median completion time of 1 h and 9 min.

Procedure and materials

Once participants enrolled in the study on Prolific, they were directed to a Qualtrics survey. Following informed consent, participants completed a demographics questionnaire. They then progressed to the primary part of the study, which was a series of affective and creativity measures. We note that some measures that participants completed were listed as exploratory in our preregistration, and thus we do not report on their results in this paper. Interested readers can find information on these questionnaires in the Supplemental Materials.

Affective questionnaires

First, participants completed the 40-item Profile of Mood States29 to measure state-level mood disturbance (negative subscales minus positive subscales). Participants rated the level (“Not at all” to “Extremely”) that they currently felt a variety of emotional adjectives (e.g., “tense”). However, as in Faul and De Brigard30, we included five adjectives (“happy,” “hopeful,” “relaxed,” “content,” and “clear-headed”) that comprised an additional general positive affect subscale, allowing for a more comprehensive indexing of positive mood (see also31.

Then, participants completed the 50-item Survey of Coping Profiles Endorsed (SCOPE27, which measures their habitual usage (1 – “Never” to 7 – “Almost always”) of 14 different ER strategies in recent months: active distraction, cognitive distraction, cognitive restructuring (reappraisal), emotional containment, emotional expression, humor, other-blame, passive resignation, problem-solving, religion, rumination, self-blame, social support seeking, and wishful thinking. Definitions for each strategy can be found in Table S2. Each strategy subscore was calculated by averaging across responses of the respective items. These SCOPE results were used as our measure of habitual ER tendencies, henceforth denoted as SCOPE Self-Report Scores. Crucially, because our study also sought to examine the correspondence between these self-reported endorsements of ER strategies and our novel ER generation task, we base our free-response coding scheme on these 14 SCOPE strategies (see Strategy coding).

Then, participants completed the 20-item Center for Epidemiological Studies: Depression Scale (CES-D32). The CES-D involves rating behaviors or feelings as occurring “Rarely or none of the time” (0) to “Most or all of the time” (3), with summed scores reflecting depression symptom severity. Typically, a score of 16 has been shown to be an appropriate cutoff for identifying individuals at risk for clinical depression32,33, although here we used depression severity scores as a continuous predictor in all our models. These symptom-level scores are denoted as Depression Severity in the remainder of our manuscript.

Following the CES-D, participants completed the 24-item joint Behavioral Inhibition Systems (BIS) and Behavioral Activation Systems (BAS) Scales as measures of motivation subtypes34. Higher summed BIS scores reflect greater inhibition sensitivity (e.g., “Criticism or scolding hurts me quite a bit”), whereas higher summed BAS scores reflect stronger approach motivation tendencies (e.g., “I go out of my way to get things I want”). Motivation was a construct of interest given that it is a correlate of anhedonic tendencies in depression. Previous research has suggested motivation to be a crucial component to both creativity and affect (e.g., “people in a sad mood… do not produce higher levels of creativity because they lack the approach motivation and behavioral tendencies”35, p. 784). Given our emphasis on approach motivation in particular, our preregistered analysis plan intended to explore only BAS scores as a predictor for ER generation patterns; hence, we do not report on BIS scores here.

ERGen: a novel ER generation task

Next, participants were directed to the main task of the study involving the generation of ER techniques. First, participants were told that they would be receiving 20 short (two-sentence), emotionally-negative scenarios in which to imagine themselves, and that they had one minute to “generate and list as many ideas that could possibly manage your negative emotions” in response to each scenario. The 20 presented scenarios were randomly drawn from a pool of 30 possible negative scenarios (see Supplemental Materials for more information, including Table S1). Thirty-second reminders appeared halfway through each scenario presentation. The one-minute time-limit was chosen in accordance with typical DT paradigms36 and following pilot testing to allow for sufficient generation while minimizing participant fatigue. While we collected valid data with this time-limit, we note that all DT tasks measuring verbal fluency are potentially sensitive to individual differences in typing speed. Although we partly controlled for this effect by ensuring participants used the same device (computers), we do not have access to the timing of participants’ keyboard clicks and thus cannot comprehensively control for this potential individual difference. Thus, differences in technique fluency (see Figure S2) could be attributed in part to factors beyond verbal fluency ability, including typing speed.

Participants were asked to be specific with each idea by using details from the scenario, but that they didn’t necessarily have to solve the problem described in the scenario. Instead, they should focus on managing their personal feelings, which could happen through internal thoughts as well as behavioral actions. Lastly, we encouraged participants to list any and all ideas they came up with in order to encourage the spontaneous and open-ended nature of the ERGen task. This spontaneity should, in theory, represent what strategies are readily accessible for the participant.

Before the 20 scenarios were presented sequentially, participants had an opportunity to practice with the scenario “You are moving this week. Your new place is inadequate relative to the old one.” The practice trial was not analyzed and was only intended for participants to clarify what to expect from the main task (e.g., timing, page format). Participants were not given any feedback on their performance in the practice trial. Following instruction and practice, participants completed the ERGen by generating techniques to each of the 20 scenarios which we then coded and modeled (Fig. 1). At the end of the study, participants were shown the 20 scenarios they previously saw and were asked to rate each scenario on several contextual variables used for exploratory analyses (see Table S1 for average ratings for each scenario). Specifically, these ratings sought to clarify the specific aspects of these scenarios that may have influenced generation beyond individual differences like creative-thinking ability and depression (see Supplemental Materials for analyses). This rating component was the final task before participants were debriefed on the nature of the study and compensated.

In summary, ERGen is a task that elucidates individuals’ repertoire profiles—a relatively under-researched aspect of ER—through DT principles, namely the use of free-response prompts that have no constraints except timing (one minute), and whose responses are examined for fluency and novelty (specifically diversity). Through ERGen, we offer one possible method by which to measure idea generation in ER in a manner that relies on qualitative data, a notable divergence from most ER research and an extension from the reappraisal-specific studies of others13,14,15,16. The hypothetical scenarios for this task were validated in prior studies as reliably eliciting negative emotions37,38, although we extend their purpose here as prompts for qualitative ER idea generation.

Creative ability

After the generation stage and before the rating stage of the ERGen task, participants completed four, one-minute rounds of the Alternative Uses Task (AUT39, which involves generating alternative uses for everyday objects as an index of creative-thinking ability. Based on the standardized list from Beaty and Johnson36, we prompted participants to generate creative uses for the following objects in this order: candle, brick, pillow, pencil. Participant responses were cleaned by the first author for sensical data (e.g., comprehensible and appropriate words) and then input into the Open Creativity Scoring with Artificial Intelligence (OCSAI) website, version 1.640. OCSAI is an open-source scoring method that leverages large language models trained on thousands of previous AUT responses and outputs originality scores (1 [very unoriginal] to 5 [very original]) that have been shown to correlate at r =.81 with human raters. These originality scores reflect the statistical uncommonness of each response: a facet of novelty within DT. For each participant, we derived the average originality score across responses for all objects. For simplicity, we refer to this average score as Creative Ability.

Strategy coding

We developed a strategy coding scheme (Table S2) based on items from the SCOPE, as well as strategy definitions provided by Beckham and Adams41 (the original questionnaire on which the SCOPE was based and validated). Additional definitions were adapted from Aue et al.42, Nolen-Hoeksema et al.43, Kugler and Kuhbandner44, and Webb et al.45. Table S2 includes definitions and example techniques generated by participants for each strategy. The codebook was refined through pilot-testing, as we discovered examples that didn’t neatly correspond with SCOPE items or the Coping Strategies Scales41 definitions. Mainly, our Passive Resignation strategy definition was amended to include uncommon examples mentioned by participants like meditation-related techniques, as we believed this was the closest strategy these examples belonged to of the 14 possible SCOPE strategies (based on their similarity to the other Passive Resignation SCOPE items). In other words, no strategy exists for mindfulness meditation, and we could not make a new one as we had to maintain the same strategy list for comparative analyses between the SCOPE and ERGen.

Two undergraduate raters trained in the emotion and ER literature coded each of the participant-generated techniques. Before cleaning, participants generated a total of 8054 potential technique responses. However, as listed in our preregistration, if one or both of the raters could not understand the technique (e.g., was non-sensical, was not an understandably finished thought), the technique was omitted from analyses. This procedure led to the final technique count of 7854. The raters agreed substantially (using standards from46 on the strategy codings at an unweighted Cohen’s κ = 0.62 [0.61, 0.63]. The first author, prior to any data analyses, then reconciled any disagreements by selecting the most appropriate option between the two potential strategies.

As mentioned in our preregistration, while our analysis plan was primarily concerned with how human raters code techniques according to strategies, we also explored an automated method of strategy coding by using OpenAI’s GPT-4o large language model. Previous work has found that GPT can successfully be leveraged to analyze text data within psychology paradigms26, but no work has yet explored this possibility in the domain of ER, largely due to the rarity of measuring ER through an open-ended approach. In addition, we wished to confirm that the human ratings were accurate by comparing them against a more objective, more automated, and less arduous method of scoring, which may further motivate future research in this qualitative coding space. We report on the full GPT-coded methods and models in the Supplemental Materials. The R scripts used to access GPT can also be found on our OSF page and were informed by prior work using a similar approach26. We note that GPT and reconciled human codings agreed at an unweighted Cohen’s κ = 0.71 [0.70, 0.72], and model results generally replicated between both coding sources (see Table S10).

Data analysis

Our preregistered study design examined two dependent variables based on the participant-generated techniques: (1) Strategy Diversity (the number of strategies represented among the generated techniques), and (2) Technique Count (the number of techniques generated per scenario). These variables correspond closely to DT principles of flexibility and fluency, respectively (see25). Across analyses, we report test statistics and p-values, and when appropriate, betas and uncorrected 95% confidence intervals (CI). Below, we list our models that were preregistered in accordance with our hypothesized notions of what variables may influence ER generation. The ultimate reasoning for the specific construction of these models was to contribute to a still-nascent theoretical framework of how DT associates with ER.

Strategy diversity

To model Strategy Diversity, we used a generalized linear mixed effects model via the glmer() function from the package lme4 (version 1.1.35.3). We specified random intercepts for Participant and Scenario. Further, because Strategy Diversity is a count variable, we specified a Poisson distribution for the model. Visualizations were made through sjPlot (version 2.8.16). We report on the following preregistered models predicting Strategy Diversity, with all continuous predictors normalized prior to regression:

$${\mathbf{Model}}{\text{ }}{\mathbf{1A}}:Strategy\;Diversity={b_1}*Creative\;Ability+{b_2}*Depression\;Severity+{\text{ }}\left( {{\text{1}}|Scenario} \right)+{\text{ }}\left( {{\text{1}}|Participant} \right)$$

To tease apart the effect of depression by its mood and motivation components, we also preregistered a model predicting Strategy Diversity that includes these variables:

$$\begin{aligned} {\mathbf{Model}}{\text{ }}{\mathbf{1B}}:Strategy\;Diversity = & b_{1} *Creative\;Ability + b_{2} *Depression\;Severity \\ & + b_{3} *Mood\;Disturbance + b_{4} *Approach\;Motivation \\ & + {\text{ }}\left( {{\text{1}}|Scenario} \right) + {\text{ }}\left( {{\text{1}}|Participant} \right) \\ \end{aligned}$$

Technique count

To model Technique Count, we used zero-inflated hurdle Poisson mixed-effects models in order to account for the large number of 0’s in our dataset47. We specifically used the mixed_model() function from the package GLMMAdaptive (version 0.9.142). The data were zero-inflated since, for each scenario, any strategy with no techniques generated was assigned a value of 0, and participants often only generated techniques for 1–3 strategies per scenario. Hurdle models fit two distinct parts: the zero-part (a logistic regression that models the probability of whether the data point is 0 or not; henceforth referred to as the “Choice” part), and the fixed-effects part (a truncated-at-zero Poisson distribution that predicts the number of techniques with a beta coefficient if it is indeed non-zero; henceforth referred to as the “Quantity” part). We report on both parts of the hurdle models throughout the manuscript. To reduce the complexity of the random effects of the hurdle model and to allow for convergence, we only included a random intercept for Participant for both the Choice and Quantity parts. Post-hoc tests and visualizations from these hurdle models were conducted using joint_tests(), emtrends(), and emmip() from the package emmeans (version 1.10.2), as well as with visualizations from ggplot2 (3.5.1).

In terms of interpretation, the Choice beta reflects the probability of generation, whereas the Quantity beta reflects the number of generations (interested readers can find more information on deriving the beta coefficients in the Supplemental Information). Importantly, effects from each the Choice and Quantity parts may not be related to the same strategies. For example, all participants, regardless of creative-ability levels, may have a similar probability of generating at least one technique (Choice) for, say, the strategy of emotional expression (i.e., participants are equally likely to choose to use emotional expression at all). However, it is still possible that more creative (compared to less creative) individuals generate significantly more techniques (Quantity) of emotional expression (i.e., creative ability predicts fluency in emotional expression once it is chosen).

We report on the following preregistered models (Models 2 and 3) predicting Technique Count, with all continuous predictors normalized prior to regression. Model 2 sought to test the same key variables as in Model 1, although now predicting specific techniques rather than overall diversity and also including Strategy Type to clarify the categorization of the raw techniques. By including Strategy Type (a factor variable with 14 levels corresponding to each coded strategy), we can answer questions such as how individuals generate techniques for Strategy A compared to Strategy B, and if these strategy-specific patterns are influenced by variables like Creative Ability and/or Depression Severity.

$$\begin{aligned} {\mathbf{Model}}{\text{ }}{\mathbf{2}}:Technique\;Count{\mkern 1mu} = & {\mkern 1mu} b_{1} *Creative\;Ability{\mkern 1mu} + {\mkern 1mu} b_{2} *Depression\;Severity{\mkern 1mu} \\ & + {\mkern 1mu} b_{3} *Strategy\;Type{\mkern 1mu} + {\mkern 1mu} b_{4} *[Creative\;Ability \times Strategy\;Type]{\mkern 1mu} \\ & + {\mkern 1mu} b_{5} *[Depression\;Severity \times Strategy\;Type]{\text{ }} + {\text{ }}\left( {{\text{1}}|Participant} \right) \\ \end{aligned}$$

Model 3 was preregistered to examine how self-reported use of strategies (responses on the SCOPE questionnaire) associated with generation patterns and to test if there was convergence across these measures. Habitual self-report scores are thus included as predictors in our models and referred to as SCOPE Self-Report Score. SCOPE Self-Report Score as a main effect itself carries no information about specific strategy endorsement scores, reflecting only the overall convergence between habitual and generated ER across strategies. However, when including Strategy Type as an additional moderator to SCOPE Self-Report Score, we can evaluate how convergence is moderated by the specific strategy in question (e.g., predicting the probability and counts of reappraisal techniques based on how often a participant reports via the SCOPE to use the specific strategy of reappraisal).

We included Depression Severity as an additional variable in this model to test if depression moderates this relationship between self-report and generation patterns (across strategies and within specific strategies). We hypothesized this moderation by Depression Severity to yield increased convergence between self-report and generation patterns because context-insensitivity is a demonstrated correlate of depression48, and thus stronger depression symptomology should show a lower willingness to experiment with (i.e., generate) instances of strategies that these participants do not often use (as measured by self-report).

We note that we based our power analysis on Model 3 (six predictors). However, given that Creative Ability emerged as a significant predictor throughout our results, we also conducted a more-complex hurdle model that extends Model 3 by including Creative Ability and its interactions, yielding 11 total predictors. We denote this non-preregistered model as Model 3S, and interested readers can find its results in the Supplemental Materials. Importantly, even when controlling for Creative Ability, the main effects and interactions involving Depression Severity, Strategy Type, and SCOPE Self-Report Score of the original preregistered model (Model 3) replicate.

$$\begin{aligned} {\mathbf{Model}}{\text{ }}{\mathbf{3}}:Technique\;Count{\mkern 1mu} = & {\mkern 1mu} b_{1} *Depression\;Severity{\mkern 1mu} + {\mkern 1mu} b_{2} *Strategy\;Type{\mkern 1mu} \\ & + {\mkern 1mu} b_{3} *SCOPE\;Self\;Report\;Score{\mkern 1mu} + {\mkern 1mu} b_{4} *[Strategy\;Type \times SCOPE\;Self\;Report\;Score]{\mkern 1mu} \\ & + {\mkern 1mu} b_{5} *[Depression\;Severity \times Strategy\;Type]{\text{ }} \\ & + b_{6} *[Depression\;Severity \times Strategy\;Type \times SCOPE\;Self\;Report\;Score]{\text{ }} + {\text{ }}\left( {{\text{1}}|Participant} \right) \\ \end{aligned}$$

Results

A correlation plot among our primary measures of interest can be found in the Supplemental Materials (Figure S3). We note that this analysis averages across the ERGen task and thus does not contain information about trial-specific generations, which is more accurately measured in the proceeding results. Full regression tables for Models 1 through 3 can be found in the Supplemental Materials as Tables S3-9.

Strategy diversity

First, we sought to answer if we could predict Strategy Diversity (the number of strategies represented among the generated techniques, per scenario) from individual creativity levels (average AUT Originality scores; Fig. 2A) and depression symptom severity levels (CES-D scores; Fig. 2B) using a Poisson-distributed, generalized linear mixed-effects model (Model 1A). We found that Creative Ability significantly positively predicted Strategy Diversity (b = 0.16 [0.11, 0.21], z = 6.25, p < .001) (Fig. 3A), such that participants who scored higher on the AUT generated a more diverse set of techniques. In other words, generated techniques corresponded to significantly more strategies as individual creative-ability increased. However, Depression Severity had no significant predictive effect on Strategy Diversity (b = 0.004 [−0.04, 0.05], z = 0.16, p =.87) (Fig. 3B).

To ensure that the effect of Creative Ability was not just attributed to prolific technique generations (i.e., that more-creative people simply generated more techniques, which happened to fall under more strategies), we added Total Technique Quantity (the sheer total number of techniques generated for each scenario, regardless of strategy assignment) as an additional covariate. Note that this more complex model was not preregistered. While we found that while Total Technique Quantity did positively predict Strategy Diversity (b = 0.18 [0.16, 0.21], z = 14.13, p <.001), Creative Ability still had a significant positive effect (b = 0.15 [0.10, 0.20], z = 6.12, p <.001).

We further sought to confirm whether the effects of Creative Ability and Depression Severity remained consistent in the presence of other emotion-related variables. In particular, in Model 1B, we controlled for the main effects of state mood disturbance (POMS scores; Figure S4) and approach motivation levels (BAS scores; Figure S5). We found that Creative Ability still had a significant positive effect (b = 0.15 [0.10, 0.19], z = 5.71, p <.001), and Depression Severity was still non-significant (b = −0.05 [−0.12, 0.02], z = −1.46, p =.14). Additionally, Approach Motivation was also non-significant (b = −0.02 [−0.07, 0.03], z = −0.84, p =.40), and Mood Disturbance was marginally significant (b = 0.07 [−0.003, 0.14], z = 1.90, p =.057). In sum, our analyses revealed that more-creative individuals generated more-diverse ER techniques (across the 14 strategies of interest), even when controlling for generative fluency and for other affective variables. In contrast, depressive symptom scores were not associated with the general diversity of strategies used.

Technique count

Creative ability and depressive symptoms as predictors of generation choice probability and quantity

We next modeled Technique Count—the number of generated techniques—based on Creative Ability and Depression Severity (Model 2). Because we were interested in how the patterns of technique generation potentially differed by the technique’s broader strategy classification, we include Strategy Type as a factor variable (with 14 levels corresponding to each possible strategy) in our models. We used zero-inflated hurdle Poisson mixed effects models, and report on both the Choice (probability of technique generation) and Quantity (number of technique generations) parts.

Choice

When predicting the probability of generating a technique (i.e., generating a non-zero number of techniques), the main effect of Strategy Type was significant (χ213 = 3730.35, p <.001). Descriptively, the most popular strategies for which participants generated (any, i.e., non-zero) techniques were problem-solving, then reappraisal, then active distraction, and the least popular ones were self-blame, then wishful thinking, then religion. Additionally, the main effect of Creative Ability was significant (χ21 = 30.64, p <.001), but not Depression Severity (χ21 = 0.36, p =.55), indicating that as Creative Ability increased, the probability of generating any technique increased. Further, there was a significant interaction between Strategy Type and Creative Ability (χ213 = 80.42, p <.001) as well as between Strategy Type and Depression Severity (χ213 = 57.66, p <.001), suggesting that both creative-thinking ability and depressive symptom levels influence the probability of generating a technique for specific strategies. We explore each interaction below.

For the interaction of Creative Ability with Strategy Type, post-hoc tests of estimated marginal trends with Sidak multiple-correction p-value adjustments revealed that as Creative Ability increased, the probability of generating a technique corresponding to the following specific strategies increased: other-blame (b = 0.34 [0.15, 0.54], z = 3.47, p =.007), social support seeking (b = 0.48 [0.34, 0.63], z = 6.61, p <.001), reappraisal (b = 0.29 [0.18, 0.39], z = 5.26, p <.001), problem-solving (b = 0.20 [0.10, 0.30], z = 4.05, p <.001), passive resignation (b = 0.25 [0.13, 0.36], z = 4.26, p <.001), rumination (b = 0.24 [0.11, 0.36], z = 3.74, p =.003), active distraction (b = 0.28 [0.17, 0.39], z = 5.04, p <.001), and humor (b = 0.44 [0.24, 0.65], z = 4.19, p =.004) (Fig. 4). These results support the notion from Model 1 that more-creative individuals have access to a more-diverse repertoire when generating ER techniques.

For the interaction of Depression Severity and Strategy Type, post-hoc tests of estimated marginal trends with Sidak multiple-correction p-value adjustments revealed as Depression Severity increased, the probability of generating a technique increased for the strategy of other-blame (b = 0.29 [0.12, 0.46], z = 3.37, p =.011). Meanwhile, as Depression Severity increased, the probability of generating a technique decreased for the strategy of reappraisal (b = −0.17 [−0.28, −0.07], z = −3.28, p =.015) (Fig. 5). These results demonstrate strategy specificity with regard to how depression symptom severity associates with technique generation.

Quantity

When predicting the number of techniques generated, the main effect of Strategy Type was significant (χ213 = 417.43, p <.001). The strategies for which the greatest number of techniques were generated were problem-solving, then active distraction, then reappraisal, while the strategies for which the least techniques were generated were wishful thinking, then cognitive distraction, then self-blame. Additionally, the main effect of Creative Ability was significant (χ21 = 4.16, p =.041), but not Depression Severity (χ21 = 0.17, p =.68), suggesting that as Creative Ability increases, participants generated a greater number of techniques. While there was not a significant interaction between Strategy Type and Creative Ability (χ213 = 10.22, p =.68), there was one between Strategy Type and Depression Severity (χ213 = 29.76, p =.005). These findings suggest that increased creative-thinking ability predicts a greater number of technique generations across strategies, but that depressive symptom levels influence the number of techniques generated only for certain strategies.

Unpacking this interaction between Strategy Type and Depression Severity, post-hoc tests of estimated marginal trends with Sidak multiple-correction p-value adjustments revealed that as Depression Severity increased, participants generated more techniques for the strategy of social support seeking only (b = 0.40 [0.16, 0.65], z = 3.21 p =.018) (Figure S6). Thus, greater depression severity associated with more generated ideas relating to seeking help from others, when this strategy was selected.

Correspondence with self-reported ER habits

Here we address the question of whether people generate techniques of strategies they self-reportedly claim to typically utilize in daily life, and whether this potential correspondence is modulated by individual differences in depressive symptoms. We report on a preregistered Model 3 focused on Depression Severity, Strategy Type, and SCOPE Self-Report Scores, although results replicate when also controlling for Creative Ability in Model 3S (see Supplemental Materials).

Choice

When predicting the probability of generating a technique (i.e., zero or non-zero technique count), Model 3 revealed a main effect of Strategy Type (χ213 = 2698.62, p <.001) and a significant two-way interaction between Strategy Type and Depression Severity (χ213 = 60.55, p <.001), as also shown with the Model 2 results. New in Model 3, however, was a main effect of SCOPE Self-Report Score (χ21 = 5.72, p =.017), which suggests a general correspondence between self-reported regulatory habits and generation patterns. This main effect of SCOPE Self-Report Score was moderated by Strategy Type (χ213 = 78.20, p <.001), with the following strategies showing significant correspondence (following Sidak p-value corrections) between generation probability and SCOPE Self-Report Score: religion (b = 0.62 [0.44, 0.81], z = 6.70, p <.001), active distraction (b = 0.26 [0.13, 0.39], z = 3.85, p =.002), and problem-solving (b = 0.20 [0.09, 0.31], z = 3.48, p =.007). Critically, a significant two-way interaction emerged between Depression Severity and SCOPE Self-Report Score (χ21 = 8.71, p =.003; Fig. 6A), suggesting that depression levels modulate the correspondence between self-report tendencies and generation, as hypothesized. Lastly, a significant three-way interaction emerged between Strategy Type, Depression Severity, and SCOPE Self-Report Score (χ213 = 96.38, p <.001). Thus, while SCOPE self-report scores do predict the probability of generating techniques on the whole, Depression Severity moderates this relationship particularly within certain strategies.

In unpacking the significant three-way interaction between Strategy Type, Depression Severity, and SCOPE Self-Report Score, post-hoc tests revealed that the following strategies showed a significant moderating effect of Depression Severity on the relationship between SCOPE Self-Report Score and generation probability: other-blame (χ21 = 4.27, p =.04), emotional expression (χ21 = 7.08, p =.008), social support seeking (χ21 = 21.68, p <.001), reappraisal (χ21 = 7.30, p =.007), problem-solving (χ21 = 26.97, p <.001), passive resignation (χ21 = 7.22, p =.007), active distraction (χ21 = 25.79, p <.001), and humor (χ21 = 5.28, p =.02). To visualize this effect, we evaluated marginal trends of SCOPE Self-Report Scores at different levels of Depression Severity, specifically at Low (−1 SD; CES-D score = 4.29), Moderate (Average; CES-D score = 16.56), and High (+ 1 SD; CES-D score = 28.83) CES-D scores. As demonstrated in Fig. 6B and C, higher Depression Severity increased the correspondence between SCOPE Self-Report Score and technique generation probability (i.e., the slope of the lines become less negative/more positive from the left to right panels) for all the above-mentioned strategies except emotional expression and reappraisal, where the opposite pattern emerged. For emotional expression and reappraisal, as Depression Severity increased, higher self-reported endorsement scores on the SCOPE were actually associated with a relative decreased probability of generating a technique for these strategies.

Quantity

When predicting the actual number of techniques generated, Model 3 replicated the main effect of Strategy Type (χ213 = 349.86, p <.001) from Model 2. This main effect was moderated by a significant two-way interaction between Strategy Type and SCOPE Self-Report Score (χ213 = 26.59, p =.014). In this added presence of SCOPE Self-Report Score, the two-way interaction between Strategy Type and Depression Severity shifted from fully significant in Model 2 to marginally significant here in Model 3 (χ213 = 21.40, p =.065). Lastly, a significant three-way interaction emerged between Strategy Type, Depression Severity, and SCOPE Self-Report Score (χ213 = 26.79, p =.013).

To further examine the significant three-way interaction, we conducted post-hoc tests to localize which strategies demonstrated significant two-way interactions between Depression Severity and SCOPE Self-Report Score. We found that only for social support seeking (χ21 = 6.48, p =.011) and active distraction (χ21 = 14.89, p <.001) did these interactions emerge, such that higher Depression Severity increased the positive relationship between SCOPE Self-Report Score and number of technique generations (Figure S7). In other words, when these two strategies were selected, individuals with a higher number of depressive symptoms showed a stronger correspondence between their self-reported tendency to use social support seeking or active distraction and the total number of techniques generated for these strategies.

Discussion

ER involves ideating specific options to manage our emotions, but what factors predict the diversity and fluency with which we generate these ideas? Using a DT task adapted for studying ER generation, we demonstrated that enhanced creative-thinking ability (originality scores on the AUT) predicted access to a diverse repertoire as predicted, evidenced by more strategies utilized per scenario and a higher likelihood to generate techniques for most of the strategies we assessed. Meanwhile, more symptoms of depression associated with altered generation patterns within select strategies (e.g., reappraisal), as well as a reduction in willingness to deviate from habitual ER tendencies during situation-specific generation. This study thus not only advances ER methodologies by revealing the rich data available from free-response paradigms, but it also informs theoretical and clinical extensions of ER by suggesting cognitive skills that could be beneficial to ER generation, including creative (original) thinking.

Creativity refers to the ability to leverage novel thoughts in useful manners49. This process is thought to drive not only the generation of uncommon uses for everyday objects (i.e., the AUT) as is most commonly assessed in the literature, but also, we argue, in any domain involving diverse thought generation. Here, we establish ER as one such domain. Previous studies have supported that ER involves generating ideas17,50, though we built on this work by integrating novel methodological (e.g., divergent thinking [DT]) and statistical (e.g., hurdle models) procedures, coding nearly double the number of strategies using both human and LLM raters, and recruiting a higher sample size across the adult lifespan. In addition, we used stimuli previously validated to evoke negative emotions as a way to elicit ideas for regulation. Through this procedure, we revealed the positive predictive effect of AUT originality scores (“Creative Ability”) on the fluency and quality of ideas generated for ER.

From a creativity theory perspective, our results signified that different facets of DT correlated with each other across tasks in different domains, lending support to the perspective that creative ability is domain-general rather than domain-specific (see51,52). In particular, the construct of novelty within DT can be measured through originality (statistical uncommonness) or through diversity (the jumping between broader categories; more commonly referred to as “flexibility” in the creativity literature). In the present study, these two facets (originality and diversity) strongly associated with one another between the AUT and ERGen, even though they are different signals of novelty [Future research interested in the correspondence between the same novelty sub-facets (e.g., ER originality against AUT originality) could adapt our ERGen task in a way that more strongly mirrors AUT instructions; that is, by asking participants to generate specifically uncommon ideas for regulation. However, this approach is not as externally valid, hence we did not implement it here.]. This result replicates findings from Weiss and Wilhelm53—who demonstrated a similar relationship using cognitive tasks—and extends them into the affective domain. Leveraging this multi-faceted profile of novelty, these results set up future work that could explore if interventions that experimentally manipulate individuals’ abilities to think originally (“outside-the-box”) have meaningful translational effects to diversifying regulatory repertoires.

Worth discussing is the ultimate adaptiveness of diverse and fluent ER generation. Taking from the ER flexibility model2, one could argue that these processes expand ER repertoires and are thus beneficial given that small repertoires are theorized to impede the ability to adapt to different contexts, particularly if combined with other risk factors (e.g., negative memory biases54. However, it is possible that strong creative ideation may not directly equate to more-effective ER given that there are other significant components of the ER process that predict efficacy. Taking stages from the extended process model1, perhaps more-creative individuals are good at generating ideas but less adept at determining when to regulate in the first place (“Identification”). This is not a component that we targeted, given that all generation was explicitly prompted. Alternatively, perhaps these individuals struggle with selecting the most appropriate technique based on their generated pool of available options (“Implementation”), particularly when equipped with such a wide and potentially overwhelming response space. Further, because we did not examine implementation, our study did not capture how well the generated techniques actually changed participants’ emotions. That said, ERGen could be adapted in a way wherein participants select from their generated ideas (implementation) and then report how successful their chosen technique was in reducing negativity or increasing positivity, akin to reappraisal work by Luo et al.8 and Waugh et al.9.

While creative-thinking ability showed rich relationships with ER generation, the main effect of depression also revealed interesting patterns of generation worth future investigation. Individuals with higher symptom severity were more likely to generate ideas corresponding to other-blame (see55) and generated more ideas of social support seeking (see56). Particularly interesting was the inhibition of reappraisal generations with higher depression severity, which converges with the literature on trait usage of reappraisal21,57. However, other literature has found that depression levels do not influence rates of spontaneous (non-instructed) reappraisal22,23,24, though our divergence may be because these studies used retrospective self-report, whereas we examined unfolding, qualitative idea generation. While these findings were fairly selective (relating to only three of 14 strategies, each in different ways), Model 3 revealed an interesting relationship between self-report and ER generation that was much less strategy-specific. Specifically, increasing depression symptoms yielded stronger convergence between self-report habits and spontaneous generation (especially for six common strategies). We interpret this effect as reduced adaptability to situation demands. In other words, individuals with greater depressive symptoms are more inclined to rely on their average ER tendencies (as indexed by self-report) across specific contexts that necessitate spontaneous ER generation48. This finding indicates that strategy experimentation (i.e., exploring strategies that individuals can ideate but do not commonly use) may provide complementary benefits to targeted skills training in depression interventions.

Taken together, these results begin to converge on a novel theoretical framework that considers ER, DT, and depression in tandem. Depression has previously been shown to impede DT ability20, perhaps due to reduced motivation and/or the narrowing effect of negative moods. However, as demonstrated in this paper, DT ability is linked with the generation of regulatory strategies in emotional situations. Thus, if depression suppresses DT, then one would speculate that depression subsequently constrains diverse ER generation. Lastly, without diverse repertoires from which to select and implement appropriate techniques (along with other risk factors), it is possible that blunted ER generation perpetuates negative emotion states by failing to appropriately regulate them. Further, it has been shown that greater resistance in changing an emotional state (a notion also referred to as emotional inertia) predicts depression (especially in emotionally laden situations58), which completes a cyclical framework between depression, impaired DT, and impoverished ER generation, and also offers a potential mechanism by which context inflexibility emerges in cases of depression48. This framework is also bolstered by the rich relationships between dynamic thought generation and mood already demonstrated in previous research. For example, it has been shown that the quality of spontaneous thoughts predicts internalizing symptoms like depression through trait-level regulatory preferences, closely corresponding to our framework59. Additionally, Nykliček et al.60 found both that previous negative mood predicted more past-focused thought, and inversely, more present-focused thought associated with more positive future mood. Adapting ER generation as a similar form of dynamic thought generation, these findings support our bidirectional framework in that lower mood can inhibit the scope of ER possibilities (which we propose to be due to poor DT), and that the scope of ER thought (e.g., if it is present-focused) also associates with better future mood.

However, it should be noted that some of our findings initially challenge this cyclical framework: depression severity did not associate with strategy diversity (against our anticipated negative prediction as based on Lougheed and Hollenstein61 and did not generally relate to lower technique generation. We speculate these null relationships to have emerged for two reasons. First, ERGen is an online task that uses hypothetical scenarios; extending it to the “real world” with stronger higher emotional stakes may elicit the anti-DT effects seen in Baas et al.20. Secondly, our work did not recruit a clinical-specific sample. ER generation in Major Depressive Disorder may support a cyclical framework posited above.

Lastly, an open question remains as to the stability of participants’ responses to the ERGen task. In other words, is DT in ER a state-level measure (dependent solely on the contextual factors of the regulatory stimulus), or is it a trait-level individual difference that is stable across instances of ER? This point closely mirrors discourse surrounding task-based creativity measures more broadly. Indeed, variables like state-level mood and motivation have been shown to strongly affect creative thinking35. For instance, Mather et al.62 cited that their creative problem-solving task represented “an individual difference with trait-like stability… [that] can also be influenced by factors such as affective state” (p. 7362). Our results suggest a similar mixture: contextual factors like scenario negativity influenced both strategy diversity and technique count above and beyond individual difference variables like depression symptomology and AUT scores. Additionally, these scenario variables were scores that were centered within-participant (i.e., ratings about scenario features were relative to each participant’s mean), which helps isolate some within-individual variability across trials. When combined with our multi-trial design and the use of Scenario as a random effect in Model 1 (see Supplemental Materials), we view these features as a strength of our study design. That said, these state generation processes may also rely on a latent individual difference with regard to generative ability or habitual strategy preferences. Analytically, although we included these trait variables in our models to test (see Model 3S), we unfortunately could not disentangle between- versus within-person variability further given that we only conducted a single assessment of the ERGen task. In other words, our data does not answer how stable (or variable) each individual is at ER generation at multiple timepoints, which is ultimately a limitation of the study. For this reason, we encourage future research to clarify the stability of the metrics from qualitative tasks like ERGen over time and across scenarios, both real and hypothetical.

Limitations and considerations

How these generation patterns extend to real-world ER remains a rich question, given that here we asked participants to imagine hypothetical events. Note, though, that the scenario stimuli we used still evoked self-referential negative affect as evidenced by our own ratings data and prior behavioral and physiological research using the same or similar scenarios37,38,63,64,65. Moreover, it is possible that certain strategies showed non-significant relations in post-hoc tests not because they were preferred less, but because some strategies are naturally harder to generate techniques for and/or are implemented more reflexively, like shutting one’s eyes in response to a scary movie66.

Secondly, these findings must be interpreted cautiously from a clinical perspective, given that we did not recruit a clinical sample, nor did we measure other clinical features like diagnostic and treatment history (e.g., medication usage). We emphasize that our metric of depression was one measuring symptom severity, not diagnosed depression. Although these scores were well-spread across the range of symptom scores and approximately half of our participants were above the questionnaire’s clinical cut-off score of 16 (Fig. 2B), we opt to interpret these scores as a continuous variable rather than a binary healthy versus above-cutoff group to avoid potentially misleading clinical interpretations and to maximize statistical power.

Additionally, the study design was presented in a fixed order, with the AUT occurring after ERGen for all participants. This ordering was purposeful because it avoids the possibility that we cued participants to think creatively before ER generation. However, the reverse direction is theoretically possible, such that completing the ERGen task influenced generating more creative uses for everyday objects. One route by which this could be true relates to the negative nature of the ERGen scenarios, which may have served as a negative mood induction prior to completing the AUT. The rich literature of mood influences on creativity has shown that negative moods do not consistently affect creative performance, but can sometimes hinder or even enhance creativity depending on task type. Specifically, in their meta-analysis, Baas et al.35 found that “negative moods tended to produce more creativity than positive moods when the task was framed as serious and performance oriented” (p. 793). ERGen may exemplify such a task, which supports a result in our Supplemental Materials that scenarios rated higher in negativity elicited greater strategy diversity. Future research should continue examining the bidirectional interplay between affect and creativity.

Another limitation to consider is that these results are correlational. Unfortunately, extant creativity inductions that could otherwise permit causative interpretations face methodological concerns, such as a lack of control groups, small effects, and potential confounding outcomes on affective processes (e.g., mindfulness trainings67,68). Longitudinal studies may better track how creative, affective, and regulatory processes develop over the lifespan to determine their relationship.

Conclusion

In sum, we report on the creative, generative component underlying ER ideation. Methodologically, our procedure demonstrates the rich opportunities available for ER research with qualitative data collection, which we argue closely mirrors situation-specific instances of regulation in everyday life. We further demonstrate the feasibility of using LLMs to code ER strategies, an important step toward automating such qualitative analyses. Clinically, we highlight the potential for creative ideation as a target in affective disorders, particularly in depression where individuals tend to utilize a more established repertoire of habitual tendencies.

Fig. 1
figure 1

ERGen task. A Participants were asked to generate as many ideas as possible in response to 20 randomly selected negative scenarios (from 30 total), for one minute each. At the end of the experiment, participants also rated various aspects of the scenarios that they were presented. B Human raters, as well as an automated method via GPT, coded each of the generated techniques according to one of 14 possible strategies based on the SCOPE questionnaire27.

Full size image
Fig. 2
figure 2

Distribution of individual-difference variables. A Creative-thinking ability (average AUT Originality scores) raw score distribution. B Depression symptom severity (CES-D sum) raw score distribution. The mean dashed line (y = 16.5) closely corresponds with the clinical cut-off score typically associated with the CES-D (y = 16), though we do not discretize our participants by this cut-off score in our analyses. In both panels, each point represents a participant, and plots depict the full range of scores possible on each measurement.

Full size image
Fig. 3
figure 3

Creative-thinking ability, but not depression, predicts strategy diversity. A Creative-thinking ability (average originality scores on the AUT) positively predicts the number of strategies represented among the generated techniques in the ERGen task, per scenario (b = 0.16, p <.001). This effect remains when controlling for the number of techniques generated, as well as mood and motivation levels. B Depression symptom severity does not significantly predict the number of strategies represented among the generated techniques in the ERGen task, per scenario. Plots depict estimated marginal means and 95% confidence intervals.

Full size image
Fig. 4
figure 4

Creative-thinking ability predicts higher probability of technique generation across several strategies. Higher creative-thinking ability increases the likelihood of generating a technique corresponding to eight of 14 possible strategies, specifically other-blame, social support seeking, reappraisal, problem-solving, passive resignation, rumination, active distraction, and humor. ActDistraction = active distraction. Error bands = 95% CI. ***p <.001, **p <.01.

Full size image
Fig. 5
figure 5

Depression predicts probability of technique generation for other-blame and reappraisal. Higher depression severity increases the likelihood of generating a technique corresponding to other-blame, though decreases the likelihood of generating a technique corresponding to reappraisal. Error bands = 95% CI. *p <.05.

Full size image
Fig. 6
figure 6

Relationship between self-reported ER tendencies and probability of spontaneous technique generation. Across all panels, Low CES-D represents 1 standard deviation below the mean (CES-D score = 4.29), Moderate CES-D represents the mean (CES-D score = 16.56), and High CES-D represents 1 standard deviation above the mean (CES-D score = 28.83). A Depression Severity significantly modulated the relationship between SCOPE Self-Report Score and technique generation probability at Moderate (b = 0.07 [0.01, 0.12], z = 2.40, p =.016) and High (b = 0.14 [0.07, 0.21], z = 3.84, p <.001) levels. In other words, higher depression severity increases the correspondence between the likelihood of generating a technique of a given strategy and the self-reported endorsement score of that strategy, though this interaction was further moderated by Strategy Type. B The three-way interaction between Strategy Type, Depression Severity, and SCOPE Self-Report Score reveals that, as depression increases, correspondence between self-report and generated ER increases for other-blame, social support seeking, problem-solving, passive resignation, active distraction, and humor. C However, for emotional expression and reappraisal, the correspondence between self-report and generated ER decreases as depression severity increases. Asterisks by the strategy name reflect significance of the interaction between Depression Severity and SCOPE Self-Report Score within that strategy. Asterisks directly on the plots in panel (B) reflect significance of the slope of the relationship between SCOPE Self-Report Score and technique generation probability for that strategy at that level of Depression Severity. ActDistraction = active distraction. EmoExpression = emotional expression. ***p <.001, **p <.01, *p <.05. Error bars = 95% CI.

Full size image