Voice quantity.

Although it seems logical that people ‘think’ before (or while) they voice, it is not that clear yet how the voice decision process develops in the mind. Until recently, the predominant focus in the empirical voice literature has been on situational and personality factors that influence the act of voicing. Early voice measures tended to focus on whether or not people voice ideas, problems, or suggestions in organizations (e.g., ‘this employee communicates his / her opinion … even if others disagree’, [1] We refer to this as voice quantity since it assesses whether or not people speak up. More recent measures of voice do include more aspects pertaining to the content of voice, yet use items that possibly combine both quality and quantity thus making it difficult to disentangle the two [29,38]. Studies show several contextual and trait variables to be related to voice, such as safety and self-esteem [39,40], leadership [35,41], personality [4], and job autonomy [42]. Cognitive and experimental research on voice, however, is scarce and insight into the cognitive processes underlying voice decisions is limited to date.

Several theoretical ideas exist on the involvement of elaborate cognition in the voice process. For example, some authors assume that voice requires elaborate cognitive processing, because it is intentional, planned, and future focused [28,29]. Others argue that voice requires a risk-reward calculus: when judgements of efficacy and safety increase, people voice; when they decrease, voice involves risk, and thus people remain silent. Factors that influence evaluations of risk and efficacy include voice climate in groups, felt psychological safety and the perceived probability of success, organizational support, and trusting relationships with voice targets [43–45]. However, even though there are many elements to consider when deciding to voice (‘how to voice, what to voice, will someone punish me, am I able to get the message across?’), that does not necessarily mean that everyone always elaborately and consciously considers all those elements. Even complex decisions such as buying a house presumably depend largely on automatic processing [46]. So even though decisions seem highly complex, they are often taken based on automated social scripts [47].

Extensive research has shown that risky decisions are often based on heuristics and biases, reflecting automatic cognitive processing [48]. Studies show that people judge whether they feel safe or at risk based on affective properties [49], which forms a much more efficient way to navigate in a complex, uncertain world than elaborate processing. People often remain unaware of their motivations for engaging in or refraining from risk taking [50–52]. Voice research indeed shows that implicit theories about the effectiveness of voice predict whether people withhold voice that might be useful for management [53–55]. Similarly, other authors suggest that decisions to voice ‘depend on general as opposed to specific effortful cognitive appraisals of context’ [56] and might ‘operate below the level of conscious, rational decision making’ [2].

When people voice frequently, it would not be efficient to use elaborative information processing every time one decides to voice. Individuals likely simply choose to engage in the behavior that felt good in the past. Speaking up or staying silent may thus be the result of learned scripts for decisions under risk, which often involve little elaborate cognitive processing. In line with this, the only study that we know of that investigated the relationship between more general cognitive resources (GMA) and voice [30] found no association. This suggests that an ability to engage in complex thinking does not influence whether people engage in voice or not. Thus, we propose that voice decisions (voice quantity) are not the result of careful cognitive elaboration, but are rather made using fast and automatic cognitive processes.

Voice quality.

A more likely part of the voice process to involve elaborate cognitive processing, is the quality of voiced content. Although pioneering voice research has focused primarily on voice quantity (frequency), more recently, scholars have started to discriminate between types of voice on the basis of content. For example, people who feel obligated to change things at work, voice more suggestions (i.e. promotive voice), people who feel safe, voice more potential problems (i.e. prohibitive voice) [2,3], and people who self-monitor voice less deviating opinions [57]. Yet, we do not know what makes voiced information, whether it be a suggestion, a problem, or an opinion, of actual importance to the organization. While more recent content-specific measures such as promotive voice [3], constructive voice [38] or prosocial voice [58] presumably also capture quality to a certain extent, they do not explicitly disentangle quality from quantity, which is, in our opinion, necessary for studying the role of cognitive elaboration. We expect voice quality and quantity to be distinct (quantity does not always lead to quality) but inter-related components of voice behavior (e.g., frequent voicers more often experience what is valued and thus learn to voice high quality over time).

We define voice quality as ‘the discretionary constructive communication of suggestions, problems, or opinions that add novel and useful information to the organization’. This relates to creativity, which is the generation of original yet appropriate ideas in any domain [59,60]. Thus, originality or novelty plays a role in both voice quality and creativity. One other way to conceptualize voice quality would thus be to refer to creative voice, or proactive creativity [61]. However, we refer to voice quality for the purpose of contrasting the quantity and quality aspects of voice in this paper. This means that compared to organizational creativity measures [62,63], the voice quality concept has a broader focus. For example, besides sharing solutions and suggestions, it is also about pointing out important problems, and besides sharing novel ideas or challenging opinions, voice quality should always be useful. Even though usefulness is in the definition of creativity, it is usually underrepresented in creativity measures. Furthermore, in the currently available measures of creativity in organizations, there is no clear differentiation between proactive creativity (self-initiated), and reactive creativity (a response to a request from the supervisor). Since we are interested in the quality of voice as a proactive behavior, such a distinction is important.

The same proactive–reactive distinction applies for a clear differentiation of voice quality from creativity measures that are used in experimental research. Experimental research on the influence of elaborate cognition and attention on creativity uses tasks [64–66] that are reactive in nature: people are requested to generate creative ideas. They are not assumed to initiate their own creative responses and goals, and they do not socially interact with others. Investigating more proactive forms of creativity, in this case, the quality of voice, is thus important because highly creative people may generate great ideas in experiments, but never share them in a social environment. We want to study the involvement of cognitive processes not only in the generation phase, as has been done in previous work, but also within the process of deciding what to share with other people. Voice quality is thus a proactive, extra-role behavior: it refers to a voluntarily offered opinion, idea, or solution to a self-discovered problem. Voice quality is thus more social, more proactive, and contextually broader than the creativity that is typically studied in previous research.

While creativity and voice quality are thus different, we assume that they are related because both require the generation of novel information, and are thus facilitated by elaborative cognitive processing. Research suggests that Working Memory Capacity facilitates the idea generation process [65,67]. Here, we argue that WMC is likely to influence the quality of voluntarily voiced suggestions, problems, and opinions not only because it allows for original idea generation (creativity), but also because it allows for developing, selecting, and communicating those suggestions, problems, and opinions that matter to the organization (voice quality).

WMC: Cognitive elaboration resource.

‘We constantly overestimate the power of consciousness in making decisions, but in truth, our capacity for conscious control is limited’[30].

The capacity for conscious control is reflected in the capacity of working memory [26], which is limited by the quantity of information it can contain. Imagine being in a secluded park with 3 children. Two of them are playing in the grass, but the third one attracts your attention because she gets in trouble climbing up a tree. Before you attend to the tree-climber, you glance at the other two, remembering their position. You keep this visual-spatial information in mind when you climb up the tree to get the third one, which allows you to quickly locate and check on the others while you are in the tree. Now imagine the same scenario, but in a petting zoo with 8 children during high-season. Most people would fail to remember the location of all remaining 7 children, especially considering the distraction of animals and other kids. The number of children to be remembered (storage capacity) and the ability to rule out irrelevant children (distractor interference), varies greatly between people: there is a capacity limit to the memory-system. Working memory holds limited information ‘online’ to serve cognition [68]. How much we can keep attentively online at a time (Working Memory Capacity or WMC) ultimately defines the possible depth of our thoughts as well as our capacity for conscious control [26,69].

Since WMC is central to effortful and elaborate cognitive processing [25] and attention control [70–73], it is important for performing any task that requires focused and elaborate thinking, such as planning, reasoning, and persistent future goal pursuit. More specifically, the number of items (storage capacity) one can retain in the visual working memory is assumed to be equal to the number of interrelationships between elements that can be kept active while reasoning [74]. The storage component of (visual) working memory is thus a key limiting factor in our ability to understand abstract relationships between novel items [75,76]. This means that WMC is related to fluid intelligence, but is nevertheless a different construct [77,78]. WMC is a more basic, specific measure of elaborative processing capacity, which is not distorted by domain-specific, cultural, or knowledge-dependent skills. Visual working memory tasks are only about the number of items one can retain within the visual attention scope, not about reading ability, mathematical ability, or knowledge from long-term memory (crystallized intelligence).

Another way to measure the involvement of WMC in any behavior that specifically focusses on the part of WMC that controls attention, is to put load on working memory such that it cannot be used to execute other tasks [70]. Because WMC is limited, once it is ‘full’, it is occupied and cannot be used to cognitively elaborate on other issues. In our studies, we therefore use both visual WMC measures as well as attentional load manipulations to study the involvement of elaborative processing in the quantity and quality of voice. We thus investigate WMC’s involvement in voice both in terms of individual differences in evolutionary basic, visual WMC, and in terms of causality. As argued above, we expect that with regard to the quantity of voice, there is little elaborate cognitive processing involved, and we therefore propose the following hypothesis:

H1—WMC is unrelated to voice quantity.

In contrast to voice quantity, we propose that the role of WMC is more important for the quality of the voiced message. In support of a link between elaborate cognitive processing and idea development, experimental research has shown that elaborate processing is important for the generation of original ideas and the execution of musical improvisation. De Dreu and colleagues [65] showed that WMC facilitates idea generation because it allows for a persistent and perseverant focus of attention on the creative task. People high on WMC are able to stay focused on the generation process for longer periods of time and generate original ideas because they consider more possible alternatives within contextual categories.

Similarly, Benedek and colleagues [64] found effects of both WM storage capacity and WM distractor interference on creative performance, indicating that idea generation is fostered by the ability to filter out distracting information and the amount of items one can retain in working memory. If thus, elaborate processing facilitates the generation of original ideas, it may also be important for the evaluation, selection, and communication of high-quality ideas, problems, and opinions. Being able to assess whether a message contains novel information presumably requires understanding of the status quo and a thorough thought process to check whether indeed, the idea, problem, or opinion is not a mere repetition of issues that were previously raised by others.

In addition to fostering the generation and communication of original ideas, we propose that cognitive processing is likely to help people select and communicate those ideas that matter in their specific context (usefulness) [79]. describes the importance of understanding and mastering ‘the context’ in order for group improvisation to lead to innovations, instead of mere chaos. According to Sawyer, the innovation process is about selecting those ideas with potential, and making incremental changes in multiple rounds of improvisations. This incremental, implementation focused, usefulness element is important for voice because voice is social and needs to help the organization, as opposed to disrupting organizational functioning. Selection and communication of ideas, problems, and opinions that are not only original, novel, and challenging, but also useful, valuable, and implementable, requires understanding and shaping of the content. This makes the ability to keep a large quantity and variety of information active in the mind (WM storage capacity) important. Furthermore, as voice is social, the ability to filter out distracting environmental noise (WM distractor interference) while focusing attention on what to say and how to say it might be needed for selecting the best idea for communication.

The arguments above suggest that WMC is relevant to voice quality because the generation, selection, and communication of messages that are both novel and useful requires elaborate cognitive processing. We thus hypothesize that:

H2—WMC is positively related to voice quality.

Study 1 tests our hypotheses by investigating individual differences in WMC whereas in Study 2, we manipulated WMC through distracting people during the selection of voice-messages.

Independent variable WMC.

The independent variable WMC was assessed by means of a delayed serial recognition task [56–59]. Participants performed a series of 96 trials. On each trial, participants were presented with eight [8] randomly selected pictures, appearing sequentially in the center of a laptop screen. Each stimulus remained on screen for 250 milliseconds (ms), extending the complete trial over a period of 2000 ms. After a series of eight stimuli, the screen went blank for 1000 ms during which participants had to keep all information in memory. For each set we assessed response accuracy with a perfect WMC score of 96 [80].

Working Memory Capacity was calculated with a linear transformation of accuracy into capacity as used by Sligte and colleagues [64,65]. WMC was calculated by transforming the percentage of correct responses into a capacity estimate between 0 and 8, correcting for chance level (the percentage of correct responses if participants randomly select a response, which is 0.5 or 50%). We used the following formula: This ultimately results in: WMC = (% correct—0.5) * (8* 0.5). Performing at chance level would thus result in a capacity estimate of 0, and perfect performance would result in a capacity of 8. The number of items participants could retain in working memory ranged from 2.33 to 6.83 (M = 5.31, SD = .89).

Dependent variables voice quality and quantity.

Following this test, participants were introduced to their team-leader with whom they performed a filler task to get acquainted. Leaders were confederates and specifically instructed and trained for the study (namely a male and a female faculty employee who were blind to the study’s hypotheses and uninvolved in the study as authors). Gender of the leaders was counterbalanced between groups. To measure voice, we provided the team (participant and leader) with an information sharing task adapted from [81]. This task is specifically useful to measure voice and initiative behaviors in the social environment, because team-members need to voice in order to get to the best solution. We specifically created a leader-follower situation because we wanted any voice from participants to be upward, discretionary, risky, challenging and extra-role: the responsibility for the task-execution was completely in the hands of the leader, and leaders had the power to distribute payment at the end of the experiment (or so we told the participants). Participants had the chance to voice, but were never specifically requested for their opinion or help during the task.

The team-leaders’ task was to find the best new dean for the faculty. There were three candidates (A, B, and C) who all had different properties, which were selected via a pre-test of 283 students on favorable and unfavorable dean-attributes. The candidate profiles were composed in such a way that based on all available information, C would be the best candidate for the job, and A and B would both be unfavorable. However, the information that was provided to the individual team-members was (as it often is in practical situations of decision making), not identical. Based on only the leaders’ information, B would be the best option, and based on the participants’ information, A would be best. The team thus needed to share and integrate all information to get to the optimal decision, C (high quality).

After carefully reading all information on the sheet, the leader said: ‘based on my information here, it is quite obvious that we should go for option B. Here is my information, confirming my decision’. If, at that point, the participant had proactively tried to find the best candidate for the job, s/he would have found out that from his/her perspective, A was the best choice. If the participant noticed this and wanted to mention it, s/he needed to challenge the leader and voice another option. The participant could now choose (1) not to voice and comply with the leader, or (2) oppose the leader and suggest another option (voice). Whether participants voiced or not, was our measure of voice quantity. Voice quality was assessed by the relative quality of the option participants voiced. As explained previously, the participant could either choose the less optimal option A, thus staying with his or her previous preference, or derive a new perspective from the leaders’ information and voice option C, the best option from the combined viewpoints. The outcome difference between these suggestions, is assumed to be the result of the evaluation and selection process for the voice message that integrates novel information, and is most valuable to the organization (or in this case the team). Voice quality was thus measured as a dichotomous variable with low (option A) and high (option C) quality.

Idea generation phase.

Our main aim was to find out whether WMC is only important in the idea generation phase as found in previous research [65], or also important in the ‘selection for communication’ phase of voice. To make sure participants would have a pool of messages to choose from, they first anonymously generated 10 problems, opinions, or suggestions for improvement of their faculty [82] in a period of 10 minutes (for the remainder of the text, we call these: ideas). We told them explicitly that these ideas would not be evaluated or used for analysis, but that this was just an anonymous practice round. We needed 10 ideas to be able to fill both conditions (No WMC Load–High WMC Load) with 5 of the participants’ own ideas to potentially voice. During generation, the experimenter digitally classified the ideas (low/high quality). The computer used these classifications for random-weighted message distribution across the WMC Load conditions so both conditions contained an equal amount of high and low-quality messages.

Practice phase with simultaneous load manipulation.

The generation phase was followed by two combined load and selection practice tasks. We designed two tasks (one simple, one more complex) in which participants needed to select items from a pool of 10 words (the 3 largest animals) or 10 sentences (5 true statements about animals) while they performed a working memory load task (counting backwards). The purpose of these tasks was to familiarize participants with the combination of counting while selecting messages for voice, limiting possibilities for technical confusion during the real opportunity for voice.

Communication (voice) phase and simultaneous load manipulation.

Following the practice phase, participants got the instructions that they would get to see a random selection of their own ideas, as well as a random selection of other peoples’ ideas. We added 5 ideas from a database of relevant ideas to each Load condition because voice takes place in a social context. In the social environment, the evaluation of the quality of one’s own ideas also depends on the evaluation of the ideas of others. These other-ideas were kept constant between participants: each Load condition contained the full quality (low–high) range of ideas as coded by two independent judges. During the ‘window for voice’, participants thus saw 10 messages from which they could select those ideas they found ‘worthy’ to voice: 5 of their own, and 5 generated by others.

We told them that they would get the opportunity to voice any idea to the experiment-leader, who would evaluate it and pay them accordingly. Voice (as opposed to the generation phase) was thus no longer anonymous and moreover, financially risky: voicing a good idea would result in a €1,- gain, voicing a bad idea in a €1,- loss. Participants were not pushed to voice any ideas, because their primary and most profitable task was to count backwards, either from 80 in steps of 2 (High Load) or from 100 in steps of 0, thus repeating the same number vocally (No Load). Participants performed two trials with 10 messages each and a random order of High vs. No Load. Thus, participants had the opportunity to voice their previously generated ideas, while performing a WMC-load task. Voicing ideas was done digitally, by selecting and sending the message to the experimenter, who would evaluate the idea and calculate payment accordingly.

Dependent variables voice quality and quantity.

Voice quantity was measured by counting the number of voiced ideas in each Load condition. Specifically, we counted how many suggestions, problems, or opinions participants sent to the experimenter (quantity) to be evaluated. Across conditions, participants voiced 2.87 ideas on average (range = 0–10, SD = 2.29). Further, whereas Study 1 measured voice quality as an objective operationalization of value and insight, we now used expert ratings of voice quality. After the experiment, all ideas from the generation phase were evaluated by three independent judges, who rated whether the ideas provided novel solutions or useful implementations superior to the status quo, on a scale from 0 (not at all) to 6 (highly) (adopted from [83]). Inter-rater reliability was reasonable for the generated quality (ICC = .659). Voice quality was the average quality of those ideas that were voiced by the participants, which was higher (M_voiced = 3.06, se = .036, _withinCI₉₅ = {2.99,3.13}) than the quality of the generated ideas (M_generated = 2.86, se = .036, _withinCI₉₅ = {2.78,2.93}), F (1,64) = 7.76, p = .007, η² = .108) Within-participant confidence intervals were calculated using the method developed by Loftus and Mason [84], as revised by Cousineau [85]. The same method will be used to report all within subjects 95% confidence intervals throughout the rest of the paper.

Independent variable WMC.

Participants first took a visual WMC test (programmed in Presentation version 17.3) which was a combination of change localization and storage tasks which has been used in previous research on visual WMC [85] . There were three different basic tasks: change localization, serial spatial memory, and simultaneous spatial memory. All basic tasks were executed with pure storage attributes as well as distractor interference resulting in 8 different types of trials. In all trial-types, participants had to store objects (circles and rectangles) in their working memory (remember them). During this target phase, objects were displayed for 250 ms in a 5x5 grid. The location of all objects should be remembered during a delay period of 500 ms after which they disappeared. Objects were green (targets) and yellow (distractors) rectangles (120x40 pixels, 0.89 visual degree) and circles (diameter 120 pixels, 2.68 visual degree). After a 500 ms delay (objects disappeared), participants either had to recall the location of all circles or localize one rectangle that changed. The test started easy (1 object), adding objects gradually over trials, and only ended when the individual capacity limit within all trial-types was reached. The test was reliable (α = .73). WMC was calculated using Cowans’ K, a capacity estimate for tasks that require participants to remember multiple items in one visual field, as opposed to the serial task we used in study 1 [95,96].

Dependent variables voice quality and quantity.

Colleagues and supervisors rated focal employee voice quantity (15 items) and quality (15 items) with a questionnaire (Tables C and D in S1 Appendix). To control whether raters understood the behavior we intended to measure was a) discretionary, proactive, and constructive and b) distinct in the sense that quantity does not imply quality and vice versa, we first explained this followed by control questions. Voice quality was meant to reflect novelty and usefulness of voiced information (suggestions, problems, and opinions). Voice quantity was reflected in the frequency of voiced information (suggestions, problems, and opinions). One voice quantity item was deleted due to poor factor loadings. Cronbach’s’ alphas were high for colleague (α _quality = .93, α _quantity = .89) and supervisor ratings (α _quality = .96, α _quantity = .91). Descriptives and correlations can be found in Table E in S1 Appendix.

Measurement model.

Our first aim was to replicate our pilot-study findings (Tables A and B in S1 Appendix) that voice quality and quantity are separate constructs. If voice quality and quantity are indeed distinct, a 4-factor (2 sources x 2 types) measurement model should fit the data better than a 2-factor (2 sources) model which represents voice as one construct. We tested this with a CFA in Mplus. As expected, the 4-factor model with a separation of quantity and quality provided a good fit (χ² = 167.07, df = 101, N = 144, p < .001, CFI = .97, TLI .96, RMSEA = .07, 90% CI {.05, .08 }, AIC = 4055, BIC = 4206). We compared this model to a 2 factor-model, with quantity and quality as one factor, and the source (colleague vs. supervisor) as separate factors, which showed a significantly poorer fit (Δχ² = 397.53, df = 2, p < .001), (χ² = 564.60, df = 103, N = 144, p < .001, CFI = .78, TLI = .74, RMSEA = .18, 90% CI {.16, .19 }, AIC = 4448, BIC = 4594). This again suggests that voice quality and quantity are distinct constructs, not only when rated by supervisors, but also by colleagues. More details and in-depth analyses can be found in the supplementary materials (S1 Appendix), factor loadings (range .74 –.95) and scale details can be found in Tables C and D in S1 Appendix.

Sample and experimental design.

An obvious limitation of this study is the use of a student sample and of simulated organizational tasks. Although we tried our best to simulate the organizational environment carefully by using an actual social interaction, leaders, evaluations, and (financial) riskiness of voice, the external validity of our two experiments may have suffered from the highly controlled settings. The findings from the two laboratory studies thus require a replication in samples of employees (which we did in study 3), but also in field experiments and in real-time business environments (e.g., a study with recordings of actual voice behavior during meetings [104].

Voice measures.

Because we wanted to show that our hypotheses hold across different types of research methods, we also used different types of voice measures, which we feel is a strength. We have measured the quantity and quality of voice differently in all studies with the intention of showing that the distinction between voice quantity and quality is not measurement (and thus method) specific. The convergence of the results across all three studies provides evidence for the validity of our theoretical distinction. Yet, we acknowledge that our measure of voice quality in study 1 is closer to decision making quality than our measures of voice quality in Studies 2 and 3. However, we aimed to explore what mentally happens when people decide to voice or not, and what they decide to voice (and what the quality of that ‘what’ is), and our voice measure in Study 1 covers the core aspects of voice even though it incorporates the originality and usefulness of voice elements differently than the measures in Studies 2 and 3.

One other point of note, is that the correlation between supervisor-ratings of voice quantity and quality was rather high (higher than for colleague ratings). This could indicate that supervisors are less able or inclined to separate the two. It could also imply that if employees choose to voice they may be more selective in what they voice to supervisors compared to what they voice towards colleagues. First, if an employee feels that s/he has something valuable or useful to say, the likelihood of voicing to a supervisor presumably increases. Second, if voice quality towards supervisors is low, negative feedback might influence quantity of voice more severely than in the case of voice towards colleagues. Furthermore, even though the relationship between supervisor rated quantity and quality of voice was relatively strong, WMC was still only related to voice quality, not to voice quantity. So even though voice towards supervisors might be riskier than voice towards colleagues, WMC is not more strongly related to how often employees voice towards supervisors than to how often they voice to colleagues. This is in line with our idea that deciding to voice is not necessarily elaborated because risky decisions rely on automaticity rather than on cognitive elaboration.

WMC measures and manipulations.

Although our results show the same pattern across three studies, we have to point out that we used different types of WMC measures in each study and that the theoretical implications might be slightly different for each study. In our first experiment, we used a visual WMC measure that may not capture only working memory capacity. Because the task requires participants to recognize whether or not an object was present in the previous series, part of the score may reflect long term memory effects on top of WMC. This might explain why the mean scores were slightly higher than in Study 3, in which we used another measure of WMC based on recalling objects’ locations, as opposed to recognizing whether an object was previously presented. This mean difference, however, could also be explained by the fact that Study 3 was an employee sample whereas Study 1 was a student sample (which is supported by the fact that standard deviations were larger in Study 3). Still, it might be the case that in Study 1, better long-term memory played a role in participants with high WMC scores voicing better quality ideas.

Ideally, in study 2, we should have measured individual differences in WMC on top of our Load Manipulation. The complexity of the design forced us to make additional tasks as short as possible and thus limited us in adding individual differences in WMC as a covariate. Furthermore, although the literature supports the idea that our manipulation in Study 2 put load the Capacity of Working Memory, this remains an assumption. Since the use of WMC requires attention control [71–73], distracting attention with another task should (partly) limit the use of WMC [65,70,105]. However, a more conservative conclusion of the results in Study 3 would be that distracting attention from voice limits the quality of those who voice rarely. Nevertheless, we found the same interaction in Study 3, indicating that this effect is likely not only driven by attention and distractor interference, but also by the storage component of WMC.

Future research.

We tested our core assumptions about the relationship between WMC and voice quality and quantity in three different designs: a between-subjects lab design, a within-subjects experimental design, and a multi-source field study. The proposed (un)involvement of WMC in quantity and quality of voice was found and replicated in all three studies. Our study design only allowed us to test (and confirm) the negative interaction between WMC and voice quantity in Studies 2 and 3. It would be interesting to replicate and extend this using multiple indicators of experience and personality. Investigating differences in WMC involvement during voice for employees whose core task involves innovation and from other (non-innovation trained) departments, between teams who had proactivity training or not [106], between individuals from teams with differing climates for voice [45], or people high and low in personal initiative, would increase in-depth insight into this interaction.

Furthermore, over time, WMC might influence voice quantity through quality. Even if voiced rarely, positive responses to high-quality voice will likely enhance voice quantity in the future. Investigating mediation processes between WMC, voice quality, and voice quantity, and moderating effects of different leadership styles would benefit our understanding of what the ideal climate for voice quality is.

Finally, it is not yet clear whether WMC facilitates voice quality through storage capacity, distractor interference, attention, or all of them. Further research on these separate functions of WMC might be beneficial to better understand how WMC aids voice. Furthermore, there are other cognitive resources that are likely related to WMC, such as inhibitory control [107]. The ability to inhibit or control automatically triggered responses could be important for voice quality in other ways than through focusing attention on voicing. For example, previous research has shown that voice is more effective when employees have the ability to regulate their (negative) emotions [108]. Because voice may sometimes occur as a response to a problem or dissatisfaction, the ability to inhibit responding until one has developed something truly useful to say, may benefit chances that voice will be valued.

Implications for practice.

The findings of our studies offer important implications for practice. Cognitive neuroscience has shown that visual WMC might be a fundamental basis for complex thinking [24,109]. Visual working memory tests are faster, cheaper, evolutionarily more basic and less culturally biased than the more general mental ability tests that are common in management literature and selection procedures. They do not require domain specific knowledge and thus do not obstruct people with for example, dyslexia or other language (due to being from a different culture) disadvantages. In terms of enhancing diversity in organizations, this might thus be a better way to assess whether people are able to engage in elaborate thinking.

Regarding our dual pathways to voice quality, there are two important issues to consider for managers (if they wish to enhance quality voice). First, some people (infrequent voicers) need cognitive resources to voice high quality. Measuring WMC before hiring people may thus be a way to select those. However, these people also elaborate to reach high-quality voice, and elaboration takes time and effort. Merely hiring high WMC people is thus not enough, there needs to be time and space to elaborate. Second, some people (frequent voicers) need to be able to practice voice in order to learn how to improvise. Since quantity of voice seems to be more automatic (not well elaborated), trying to influence this process with rational arguments might not be very effective. People either voice often because it is in their personality, or because in their team, there is an implicit feeling that voice is safe. If people in a team generally voice rarely, it may be more useful to try to enhance voice by focusing on the implicit cues the team leader is sending, than explicitly telling people they can always share their concerns.

Theoretical contributions.

The first contribution of this study is the distinction between voice quality and quantity. It opens up new avenues for studying the multidimensionality of voice behavior and emphasizes the difference between a decision to say something and the quality of what is said. This focus on content relates to both the work on creativity, such as the combination of originality and usefulness for truly innovative ideas [110], and to voice research, such as the (complexity) differences between suggestions, opinions, and problems [2], and the differences between promotive versus prohibitive voice [29]. This paper strengthens the claims of those studies that it is important to consider the content of voice if we want to know more about how effective voice is generated, selected and communicated.

Our second contribution is the investigation of the involvement of cognitive resources in voice behavior. Morrison [2] called for more research in the cognitive elaboration versus automaticity area, which we answered by challenging the idea that voice behavior in and of itself is the result of a complex cognitive calculation. We attempted to sharpen the hypotheses about the involvement of elaborate cognitive processing in voice by showing that the arguments of Chiaburu and colleagues [28] about cognitive elaboration hold, but only for the quality of voice and especially when employees don’t voice often. Our results are also in line with the idea that choosing (not) to voice depends on affect and implicit theories about voice [53]. Our findings thus reconcile both views by showing that indeed, voice quality (sometimes) requires elaborate processing, whereas voice quantity relies on automatic processing. The results of this paper emphasize the importance of a cognitive-processes stream in the voice literature. Automaticity in a self-initiated, discretionary proactive behavior such as voice may seem like a paradox, but often, cognitive elaboration might be less important than we think.

Third, we contributed two different experimental designs (within- and between subjects), which makes a wide range of causal research questions about individual differences as well as situational differences about voice and initiative possible. Although we did not use our first lab-setting for causal purposes, the design does offer that possibility [111]. Since experimental research in voice and proactivity literature is still scarce, we hope that this helps other researchers to tackle causal research questions concerning proactive behavior.