2.1 Thought control

Thinking too much about threatening or negative issues leads to increased accessibility of negative memories (Lyubomirsky et al., 1998) and thus to an increase in negative or even depressed mood (Morrow & Nolen-Hoeksema, 1990). It has also been shown to reduce problem solving effectiveness (Lyubomirsky & Nolen-Hoeksema, 1995) and to decrease self-esteem (Roberts et al., 2016). Given that individuals with lower self-esteem also tend to be more attentive to negative information and ruminate more (Tafarodi et al., 2003), stopping negative thoughts could potentially be a mechanism to prevent a vicious cycle of negative thinking, lower well-being, and a negative self-concept.

Thought suppression is a component of thought control and describes the phenomenon of stopping particular thoughts (Anderson & Huddleston, 2012; Wells & Davies, 1994). One way to study thought suppression and its underlying inhibitory processes is the TNT paradigm (Anderson & Green, 2001). According to this paradigm, it is generally possible to control and stop the retrieval of unwanted thoughts and memories: After learning a list of word pairs, participants are asked to either facilitate (think) or stop (no-think) the respective connection in their minds. It is typically found that “think” items are remembered better compared to baseline items (control condition; word pairs are no longer presented after the initial learning phase), and “no-think” items are remembered more poorly. This effect shows that people are generally able to functionally control and extinguish memories from their minds, and that suppression intentions can lead to actual forgetting (Anderson & Huddleston, 2012). This form of retrieval suppression must be distinguished from classic thought suppression experiments, such as the “white bear” paradigm (Wegner et al., 1987). Being instructed not to think about something (e.g., white bears) often produces opposite effects. However, the differences between the two tasks might be explained by the underlying mechanisms: Whereas the instructions in the white bear paradigm already include the to-be-avoided thought (i.e., the white bear), instructions in the TNT task are more subtle and therefore make it possible to stop the respective thought before it enters the mind (for a detailed review, see Anderson & Huddleston, 2012).

However, studies using the standard think/no-think paradigm in the laboratory might underestimate thought suppression effects in real life (see also Hertel et al., 2012). In the controlled laboratory setting, people try to suppress because the experimenter tells them to, without expecting positive consequences for themselves. In everyday life, people’s motivation to suppress personally unwelcome memories should be higher due to stronger suppression motives (Anderson & Hanslmayr, 2014): Keeping down negative or threatening thoughts can help regulate emotions and affective reactions (Engen & Anderson, 2018; Erdelyi, 2006; Gagnepain et al., 2017; Nørby, 2015), reduce feelings of dissonance (Shu & Gino, 2012), maintain beliefs and attitudes (Waldum & Sahakyan, 2012), forgive others (Wilkowski et al., 2010), maintain attachment and relationships (Freyd, 1996), and preserve a positive self-perception (Sedikides & Green, 2009). All of these main functions of forgetting share the aim of maintaining a positive, coherent self-image in everyday life (Fawcett & Hulbert, 2020). Suppressing unwanted thoughts thus constitutes an important protective mechanism, not only to preserve task focus, cognitive flexibility and a positive emotional state (e.g., Anderson & Hanslmayr, 2014; Engen & Anderson, 2018; Fawcett & Hulbert, 2020; Niessen et al., 2020; Nørby, 2015), but also to protect self-esteem.

2.2 Self-esteem

Self-esteem comprises two key elements – self-competence and self-liking. Self-competence is defined as the valuative experience of oneself as a causal agent, and can be regarded as an overall conception of oneself as an effective source of power. Self-liking is defined as the valuative experience of oneself as a social object and describes an overall sense of social worth (Tafarodi & Swann, 2001). Although most people exhibit a stable level of self-esteem over time (Trzesniewski et al., 2003), individual life experiences can bring about changes and undermine stability (Conley, 1984). Self-esteem can even vary from day to day (e.g., Neff et al., 2012): Everyday experiences, as well as the accessibility of related thoughts and memories, are likely able to influence self-esteem. Consequently, deciding to deliberately control one’s thoughts should be important for protecting self-esteem.

Self-enhancement theory suggests that people are generally motivated to protect and enhance their self-esteem (Sedikides & Green, 2009; Sedikides & Gregg, 2008), and research shows that people tend to remember past situations in a generally positive manner (Demiray & Freund, 2014; Wilson et al., 2009). Demiray and Janssen (2015) found that higher self-esteem is associated with feeling closer to positive memories and that this effect is mediated by (a) the importance of the underlying memories and (b) rehearsal. This suggests that thinking more about positive experiences enhances self-esteem. By analogy, thinking less about negative experiences should have the same effect. The need for self-enhancement should motivate individuals to generate corresponding suppression intentions. As self-esteem-threatening thoughts can affect self-competence (e.g., failures, negative feedback) as well as self-liking (e.g., conflicts, own misconduct), both facets of self-esteem should be involved. We propose that the intention to suppress thoughts in daily working life is associated with better self-liking and self-competence ratings, because it prevents employees from investing too much energy in detrimental, self-doubting thoughts (leading to higher self-liking) and allows them to quickly refocus and return to the current work task (leading to higher self-competence).

3.1 Negative affect

When unwanted intrusive thoughts are not stopped, this ongoing rumination might lead to a permanent focus on these negative thoughts and ideas. Repetitive, negative thoughts and rumination can cause sadness, fear and anger and have a strong association with negative affect (see Thomsen, 2006, for a review). Negative affect is known to be negatively related to self-esteem (e.g., Lorr & Wunderlich, 1988; Watson & Tellegen, 1985): Dealing with negative affect leads to a generally negative focus and a devaluation of the self (particularly in terms of self-liking). Moreover, consequences such as narrowed attention can affect self-competence ratings. Stopping inappropriate and unwanted thoughts and memories can thus enable individuals to regulate their emotions (Engen & Anderson, 2018), refocus and set themselves back to a neutral point. We therefore expect that trying to suppress negative thoughts leads to less negative affect, followed by better self-liking and self-competence ratings.

3.2 Task focus

Task focus refers to the extent individuals orient their behaviors towards the goal of meeting job demands (Brown et al., 2005) – that is, paying attention to stimuli that are relevant to current goals while ignoring or minimizing the influence of possible distractions (Astle & Scerif, 2009). These distractions can be external factors (e.g., noise), but also intrusive thoughts. They should threaten task focus and ultimately spill over to self-esteem ratings for two reasons: First, negative thoughts and rumination decrease attentional control insofar as anxious individuals allocate most of their attentional resources to their worries (Eysenck & Calvo, 1992). This impairs processing efficiency and leads to reduced performance (Eysenck et al., 2007). Second, disruptive thoughts are associated with other deficits in executive functioning and can cause subsequent declines in cognitive task performance (van Vugt & van der Velde, 2018). Task focus and performance can, in turn, affect self-esteem ratings. Positive correlations between performance and self-esteem have been demonstrated repeatedly (for meta-analytic findings, see Judge & Bono, 2001). In our reasoning, the intentional suppression of unwanted thoughts improves task focus, leads to better performance, and enhances self-esteem ratings. We further propose that both facets of self-esteem are involved. Even though perceived performance is more closely linked to self-competence in terms of content, the resulting level of self-satisfaction might be interpreted more globally and spill over into self-liking.

6.1 Subjects

The study was conducted with a sample of 167 employed persons, who were recruited through personal acquaintance, circular letters, and telephone calls. Participation was voluntary and participants received 50 Euros as remuneration. Due to the requirements of our statistical analyses, we only included participants who completed the think/no-think procedure, filled out at least two experience sampling questionnaires,¹ and reported disruptive thoughts at least once. Therefore, our final sample comprised 143 persons (82 women, 61 men) aged between 18 and 63 years (M = 35.77, SD = 11.69). Our sample had an above-average level of education, as over 50% held a university degree, and 25% had a management position.² Participants primarily worked in office jobs (e.g., in public administration), for at least 20 h during the experience sampling week, were able to speak German at native speaker level, had normal or corrected-to-normal vision and provided informed consent.

6.2 Procedure

The study consisted of two main parts. First, participants attended an individual laboratory appointment where they performed the TNT task and received brief training and technical advice for the experience sampling. Second, data on thought suppression intentions in everyday working life were collected during an average working week (five consecutive days). Participants were fully debriefed after finishing the final experience sampling questionnaire.

6.3 Think/no-think task

In general, our work followed the procedure and materials from Streb et al. (2016). However, we reduced the number of word pairs and slightly adapted presentation times. In total, we used 58 weakly related, neutrally valenced German word pairs (see Appendix S1). Ten word pairs were used as fillers and for training purposes, while the remaining 48 were split into three groups and rotated across experimental conditions (baseline, think, no-think) and across subjects. All word pairs comprised one cue (left) and one response word (right). All words were nouns, and the order within a word pair was not changed during the experiment (i.e., cues remained cues and response words remained response words). Throughout the experiment, each word pair was presented in the center of a computer screen (for exact presentation times, see descriptions of the phases below) against a white background using E-Prime 2.0 software (Psychology Software Tools Inc., Sharpsburg, MD) and separated from the following word pair by the brief presentation of a fixation cross as the inter-stimulus interval (ISI).

The TNT task is structured into four phases (Anderson & Green, 2001; Benoit & Anderson, 2012). In the first phase (learning), participants studied all 58 word pairs in random order. Each word pair was presented in black for 6000 ms (ISI: 600 ms).

In the second phase (recall), participants were asked to recall the response words and to say them out loud. Therefore, each cue word was presented for up to 4000 ms (ISI: 600 ms) in a pseudo-randomized order (same order for all participants). As the correct answer was presented immediately following the participant’s response for 1 s, participants had the opportunity to check the correctness of their answer. If they remembered less than 50% of the critical word pairs, the recall phase was repeated. Finally, we assessed which of the word pairs had been learned by using the same procedure (criterion test, presentation time up to 3300 ms, ISI: 1100 ms) without providing feedback. This procedure limited further uncontrolled learning.

The third phase (think/no-think) constituted the main part of the task. Cues were presented for 3000 ms (ISI: 1000 ms) and participants’ tasks varied depending on the word color: For think cues (appearing in green), participants had to further rehearse the related response word in silence. No-think cues (appearing in red) asked participants to immediately stop any thoughts about the response word. Participants were instructed to only focus on the presented no-think cue and to block out any related thoughts without distracting themselves (direct suppression instruction, see Benoit & Anderson, 2012). We trained participants in this procedure using filler items. During the actual think/no-think phase, each of the 16 think and 16 no-think cues was presented 12 times. The order of presentation was randomized, but repetition was only allowed after all cues had been presented. To prevent computer eye strain and tiredness, we included three short breaks (45 s).

In the final phase (final recall), participants were shown all cue words (baseline, think, no-think) again in a pseudo-randomized order (same order for all participants). The words were presented for up to 3300 ms (ISI: 1100 ms) in black font and participants were asked to recall the formerly associated response word out loud, irrespective of the instructions in the think/no-think phase. Although this phase was meant to be unexpected, participants often assumed something like a final recall test. However, no one indicated he or she had made any effort to violate the instructions to keep remembering the no-think word pairs.

6.4 Experience sampling

The experience sampling data were collected using a self-programmed Android smartphone application. Participants could either use their own smartphone or borrow one from the laboratory. All participants received advice and training regarding the data collection procedure. The experimenter explained the concept of thought suppression and provided technical information. Moreover, participants were informed that they should start to answer the questionnaires as soon as possible after hearing signals from the application, not interrupt the process, and not interpret the signals themselves as distractions. Before leaving the laboratory, they set up the application with their individual data (anonymous personal code, participation week, working hours, lunch break) and completed a demo questionnaire. All participants received written instructions and contact information.

During the participation week, the application gave signals consisting of short beeps or vibrations three times a day. Signals were sent randomly during participants’ individual working hours, but not during their lunch break, and with a minimum time interval of 80 min. Following a signal, participants had up to 10 min to initiate responding. To minimize participants’ time expenditure and enhance response rates, we refrained from using complex scales and mainly applied one-item questions.

At the beginning of each questionnaire, we asked about participants’ task focus (“At the moment, it is easy for me to concentrate”) and their negative affect (e.g., upset) using a shortened five-item version of the Positive and Negative Affect Schedule (PANAS; MacKinnon et al., 1999). We further assessed self-esteem using two items (self-competence: “At the moment, I am highly effective at the things I do”; self-liking: “At the moment, I am very comfortable with myself”) from a revised version of the Self-Liking/Self-Competence Scale (SLCS-R; Tafarodi & Swann, 2001). Subsequently, participants were asked whether they had experienced any disruptive or distracting thoughts during the past 30 min (“Have you had to think of something that disturbed and distracted you at work in the last 30 min?”; coded as 0 for “no” and 1 for “yes”). If they responded with “yes,” they were asked about the extent (“How much did you feel disturbed by these thoughts?”), and whether they had intentionally tried to not think about, suppress or forget these thoughts (“Did you say something to yourself like ‘I do not want to think about it anymore’ or ‘I want to forget that’?”; coded as 0 for “no” and 1 for “yes”). To facilitate the response process, all items (except the dichotomous questions) were answered on 5-point scales ranging from 1 = not at all to 5 = extremely.

6.5 Statistical analyses

To analyze the results of the TNT task, we only used conditionalized data (i.e., we only included word pairs for which the test at the end of the recall phase had shown a correct response). To check whether suppression generally worked (manipulation check), we used a mixed-design analysis of variance (ANOVA) with response condition (baseline, think, no-think) as within-person factor and counterbalancing condition (A, B, C) as between-person factor. To obtain a measure for individual suppression ability, we used data from the final recall and subtracted baseline from no-think items. Higher values should thus indicate higher suppression abilities. As recommended by Anderson et al. (2004), we z-normalized all values within the participants’ counterbalancing group to ensure that our results did not rely on differences between the word pairs.

As participants completed multiple questionnaires, our data are hierarchically structured. Questionnaire data (Level 1) are nested within participants (Level 2) and must be analyzed from a multilevel perspective (e.g., Nezlek, 2001; Schwartz & Stone, 1998). We used R software (http://r-project.org) with the packages lme4 (Bates et al., 2015) and mediation (Tingley et al., 2014) for subsequent analyses. Level 1 variables were person-mean centered to consider situational fluctuations (Nezlek, 2012), whereas Level 2 variables were grand-mean centered. Between-level variance was checked by calculating intraclass coefficients (ICC₁) for all dependent variables. The ICC₁ values were 0.41 for negative affect, 0.20 for task focus, 0.36 for self-liking, and 0.26 for self-competence. Thus, there was sufficient between-level variance to justify multilevel modeling. We further applied the two-step approach to causal mediation analysis documented by Imai et al. (2010) and Tingley et al. (2014) to test for mediation and moderated mediation. In the first step, the mediator is predicted using the predictor (moderated mediation: predictor, moderator, and interaction between them). In the second step, the outcome is predicted using the predictor and the mediator (moderated mediation: predictor, moderator, interaction between them, and mediator). Finally, to test for overall mediation, the mediation analysis is run using quasi-Bayesian Monte Carlo simulations (moderated mediation: separate overall mediation analysis for high [+1 SD] and low [−1 SD] levels of the moderator). This form of causal mediation is superior to previous mediation approaches, because it overcomes problems such as dependence on specific statistical models and restrictive assumptions (cf. Pearl, 2014). Details on the underlying statistical theory are provided by Imai et al. (2010, see also Tingley et al., 2014).

7.1 Think/no-think task

In most cases (91%), participants reached the 50% criterion and therefore passed the recall phase without repetition. In the subsequent criterion test, the mean recall rate was 0.82 (SD = 0.13). The individually correctly remembered word pairs served as basis for all further analyses. Final recall rates for the previously learned word pairs were .97 (SD = 0.06) for think, 0.84 (SD = 0.19) for no-think, and 0.94 (SD = 0.08) for baseline words. We found general suppression-induced forgetting effects, as there were significant differences between conditions, F(1,193) = 58.26, p < 0.001, η_p² = 0.29. Contrast analyses showed that no-think items were recalled significantly worse than baseline items (mean difference = −0.11, F[1155] = 45.51, p < 0.001, η_p² = 0.23). This provides evidence that participants were able to actively suppress word associations for no-think items, leading to lower final recall rates for these items. Furthermore, think items were significantly better recalled than baseline items (mean difference = 0.03, F[1155] = 18.40, p < 0.001, η_p² = 0.11), indicating that reinforcing the connection during the think/no-think phase led to further learning. Consequently, the difference between think and no-think items was also significant (mean difference = 0.14, F[1155] = 90.51, p < 0.001, η_p² = 0.37). Nevertheless, it should be noted that the actual differences between conditions were relatively small, at least partially due to ceiling effects. There was no main effect of counterbalancing, F(2,140) < 0.01, p = 0.997, η_p² < 0.01, which demonstrates that forgetting effects did not simply rely on the memorability of specific word combinations.

7.2 Experience sampling

As reported above, we only included participants who completed at least two of a total of 15 experience sampling questionnaires. In total, our final sample completed 1436 questionnaires, leading to an average completion of 10.04 questionnaires per participant (SD = 3.10, range = 2–15). Hence, participants answered about two-thirds of all possible questionnaires (completion rate = 0.67).

Participants reported distractive thoughts in the past 30 min in 563 questionnaires (mean rate = 0.39, SD = 0.21, range = 0.08–1.00). These questionnaires also provide the Level 1 database for the following analyses. In about a third of all occasions when participants reported experiencing distractive thoughts, they further decided to intentionally suppress them (mean rate = .36, SD = 0.35, range = 0.00–1.00).

7.3 Hypothesis testing

In Hypothesis 1, we assumed that the intention to suppress unwanted thoughts is positively associated with both subsequent self-liking and self-competence. We conducted multilevel regression analyses, but did not find any support for the assumed relationships. Contrary to our expectations, the intention to suppress was negatively associated with self-liking (γ = −0.15, SE = 0.07, t[545] = −2.34, p = 0.020) and was not associated with self-competence (γ = 0.06, SE = 0.06, t[552] = 0.90, p = 0.370). Controlling for the extent of the disruption did not modify either these or the following results. Details of the analyses are provided in Table 3.

Hypotheses 2a and 2b referred to our propositions that a) negative affect and b) task focus mediate the association between the intention to suppress unwanted thoughts and subsequent self-liking and self-competence ratings. To test our assumptions, we used the causal mediation approach by Imai et al. (2010), which calculates quasi-Bayesian Monte Carlo simulations (1000 simulations). With regard to negative affect, we found a significant indirect effect on self-liking (indirect effect = −0.06, 95% CI [−0.11; −0.02], p < 0.001) as well as on self-competence (indirect effect = −0.05, 95% CI [−0.09; −0.02], p < 0.001). However, no significant effects emerged for task focus (self-liking: indirect effect = −0.03, 95% CI [−0.08; 0.01], p = 0.112; self-competence: indirect effect = −0.07, 95% CI [−0.15; 0.01], p = 0.110). Thus, results show that negative affect mediates the unexpected negative association between the intention to suppress and self-esteem ratings arising with respect to Hypothesis 1. Hypothesis 2a was therefore confirmed, whereas Hypothesis 2b was not.

Hypothesis 3 tested whether the association between the intention to suppress unwanted thoughts and (a) negative affect/(b) task focus was moderated by suppression ability. The analysis revealed a significant interaction effect between the intention to suppress and suppression ability on negative affect (γ = −0.14, SE = 0.05, t[550] = −2.73, p = 0.007). To interpret this finding, we plotted the interaction for high (one standard deviation above average) and low (one standard deviation below average) levels of suppression ability (Aiken & West, 1991; Preacher et al., 2006). The plot showed that individuals with higher suppression abilities were able to maintain their levels of affect following suppression intention (see Figure 2, simple slope = 0.04, z = 0.57, p = 0.567), whereas individuals with lower suppression abilities experienced an increase in negative affect when they tried to suppress disruptive thoughts (simple slope = 0.30, z = 4.25, p < 0.001). The direction of the interaction was in line with Hypothesis 3a. However, the pattern of absolute trends deviates from our expectations in its orientation (we expected a decrease for individuals with higher suppression abilities and maintenance for individuals with lower suppression abilities). Therefore, Hypothesis 3a was only partially supported.

We calculated the same interaction for task focus as an outcome, but failed to find a significant effect (γ = −0.01, SE = 0.09, t[563] = −0.06, p = 0.949). Hypothesis 3b was therefore not supported.

In Hypothesis 4, we specified our final moderated mediation models: Ability to suppress should moderate the mediation of the intention to suppress unwanted thoughts via negative affect (Hypothesis 4a) and task focus (Hypothesis 4b) on self-liking and self-competence. We conducted our analyses using quasi-Bayesian Monte Carlo simulations, in line with Tingley et al. (2014). The mediation effects were calculated for two levels of the moderator: one standard deviation below and one standard deviation above the mean (1000 simulations each). We succeeded in finding a conditional indirect effect for negative affect: When individual suppression ability was low (−1 SD), the indirect effect of suppression ability via negative affect on self-liking was significant (indirect effect = −0.12, 95% CI [−0.19; −0.05], p < 0.001). In contrast, when suppression ability was high (+1 SD), the indirect effect was not significant (indirect effect = −0.01, 95% CI [−0.06; 0.04], p = 0.640). The same pattern of results emerged for self-competence: We found an indirect effect for individuals with low suppression abilities (indirect effect = −0.10, 95% CI [−0.16; −0.04], p < 0.001), but not for individuals with high suppression abilities (indirect effect = −0.01, 95% CI [−0.06; 0.03], p = 0.640). Hence, Hypothesis 4a was confirmed for both facets of self-esteem. In order to test Hypothesis 4b, we reran the analyses with task focus as mediator, but did not find a significant indirect effect either for self-liking (low suppression ability: indirect effect = −0.04, 95% CI [−0.11; 0.03], p = 0.288; high suppression ability: indirect effect = −0.04, 95% CI [−0.10; 0.03], p = 0.210) or for self-competence (low suppression ability: indirect effect = −0.07, 95% CI [−0.20; 0.06], p = 0.290; high suppression ability: indirect effect = −0.07, 95% CI [−0.19; 0.04], p = 0.210). Hypothesis 4b was therefore not confirmed.

8.1 Theoretical and practical implications

Our study contributes to the existing research in several ways. Many studies have already examined individual ability to suppress intrusive thoughts in the laboratory (e.g., Anderson & Green, 2001; Levy & Anderson, 2012). While research on the underlying neural brain mechanisms using imaging techniques is flowering (e.g., Benoit & Anderson, 2012; Engen & Anderson, 2018), studies on consequences in everyday life are rather scarce. In the present study, we successfully replicated existing laboratory findings and combined them with experience sampling data from the work context. We also examined the consequences of thought suppression ability: We showed that affect seems to be impaired by the inability to control intrusive thoughts, and that this effect spills over to self-esteem. As such, we also contribute to research addressing underlying forgetting motives.

From an applied perspective, our findings suggest that it would be useful to support employees’ thought suppression abilities. In light of studies showing that successful thought suppression is related to suppression practice (Anderson & Green, 2001; Hulbert & Anderson, 2018), psychoeducation and trainings might be helpful to buffer employees’ emotional well-being and self-esteem against negative thoughts.

8.2 Strengths, limitations, and future research

A notable strength of the current study lies in the diversity of methods employed. We combined controlled laboratory data with data from everyday working life from the same persons, allowing us to transfer and analyze the findings and their consequences in a different context. Moreover, our sample size was relatively large compared to other studies using the TNT paradigm. This provided a solid basis for subsequent analyses, thus increasing the generalizability of the findings.

However, some limitations of the present study also need to be mentioned. First, as the predictor, mediator and outcome variables were assessed simultaneously, we cannot rule out inverse effects between these variables (e.g., lower task focus might trigger suppression intentions). We tried to overcome this problem by asking about the predictor retrospectively (“in the last 30 min”), whereas the other variables referred to the moment of questionnaire completion. Moreover, to counteract common method bias (Podsakoff et al., 2012), our moderator variable was objectively assessed using the TNT paradigm.

Second, we asked participants about their intentions to suppress unwanted thoughts, but not about their success. Instead, we used individual thought suppression ability as an indicator. In terms of avoiding common method bias, this is preferable to adding another self-report variable. Nevertheless, information about the outcome of each instance of suppression intention would have yielded more precise findings.

Third, we applied direct suppression instructions during the TNT task in contrast to thought substitution instructions. In direct suppression instructions, participants are told to not substitute the to-be-suppressed items with other words. Our goal was to keep the laboratory conditions as consistent as possible in order to get an exact measure of individual thought suppression ability. However, research shows that thought substitution is also very effective (e.g., Hertel & Calcaterra, 2005) and might be a more realistic way to stop thoughts in everyday life.

Fourth, in the final recall, participants were shown the original cues (same probe). Anderson and Levy (2007) argue that participants might generate diversionary thoughts during the think/no-think phase, which contaminates the actual thought control ability measure (see also Hertel & Calcaterra, 2005). We tried to curb this effect by applying direct suppression instructions, but using more indirect forms of final recall (i.e., independent probe; Anderson & Levy, 2007) would be preferable. Also, using intrusion ratings during the think/no-think phase (Levy & Anderson, 2012) might be a promising alternative way to measure individual suppression ability.

Fifth, our study’s generalizability might be restricted due to the above-average educational level of our sample. It seems likely that employees with high-status jobs need to deal with more complex information at work and that their tasks require more task focus. Blocking out intrusive thoughts might therefore be an automatic, well-practiced mechanism and not always be recognized as such. However, examining persons who deal with distractive thoughts on a regular basis is also the most appropriate way to address our research questions.

Finally, there might be further aspects and processes influencing the associations between our study variables which we did not cover. In future studies, it would be interesting to investigate the specific content of the unwanted thoughts. Information about further workplace characteristics could also contribute to a better understanding of the whole process.