The Field Guide › Paper
Directly resolves a long-standing ambiguity that Kish-Gephart et al. (2009) and much of the fear-and-silence literature already in this corpus leaves open: are voice and silence genuinely opposite ends of one continuum, such that whatever increases one automatically decreases the other, or are they functionally independent behaviours with their own separate drivers? The literature had split roughly evenly, with some studies finding voice and silence negatively correlated and others finding them positively correlated, with no theoretical account of why. This paper resolves the ambiguity by drawing on the behavioural activation system (BAS) and behavioural inhibition system (BIS) distinction from personality and motivation research: two functionally independent, biologically grounded regulatory systems, one appetitive and approach-oriented (associated with hope, enthusiasm, and pursuing rewards), the other aversive and avoidance-oriented (associated with fear, vigilance, and avoiding threats). Voice is theorised as a prototypical BAS response, an approach behaviour aimed at achieving a desired future state, while silence is theorised as a prototypical BIS response, an avoidance behaviour aimed at preventing self-relevant harm. Because voice and silence sit on genuinely separate regulatory systems, the paper predicts, and finds, that they have different key predictors: perceived impact, whether speaking up seems likely to actually change anything, predicts voice more strongly than silence, while psychological safety, whether acting seems risky, predicts silence more strongly than voice. One conceptual clarification worth noting: voice and silence are still opposites at the single-issue level (raise a concern or don't, in the moment), but become independent once aggregated across issues or time, meaning a person can score high on both (raising some concerns while actively suppressing others) or low on both, which reconciles the field's conflicting intuitions. The paper also directly corrects a common miscategorisation: prohibitive voice (flagging a problem or risk) has sometimes been treated as an inhibition-oriented behaviour because its content is about preventing harm, but the authors argue the act itself is an approach behaviour, taking on personal risk to protect the group, and their data bear this out. Study 1, a meta-analysis of 162 papers (voice alone drew on over 50,000 respondents against roughly 12,500 for silence, an imbalance the authors call out explicitly as an 'asymmetrical treatment' worth correcting), found the corrected voice-silence correlation was weak (r = -.15), well under the threshold usually used to judge two constructs distinct; that perceived impact explained 67% of voice's predicted variance against psychological safety's 28%; that psychological safety explained 85% of silence's predicted variance against perceived impact's 11%; and that silence related to burnout dramatically more strongly than voice did (silence explained 92% of the shared predicted variance in burnout, voice just 8%). Study 2, a six-month interval-contingent panel study, constructively replicated all of this at both the aggregated person level and the monthly, time-lagged level, and directly tested and ruled out reverse causality (voice and silence did not predict subsequent perceived impact or psychological safety, supporting the proposed causal direction). A third, unpublished replication in the paper's online appendix held even after controlling for people's dispositional BAS and BIS temperaments directly, suggesting the situational effects aren't just proxying for stable personality traits. The clearest practical implication complicates a great deal of standard psychological-safety-first advice: building safety alone may not be enough to increase voice, since voice tracks perceived impact more closely, and conversely, giving people more visible influence over decisions may not be enough to reduce silence on genuinely risky topics, since silence tracks safety more closely. The paper's suggestion is that organisations likely need two different kinds of intervention, not one: impact-generating mechanisms like quality circles or visible follow-through on suggestions to cultivate voice, and safety-generating mechanisms like anonymous channels or protection from punishment to reduce silence, rather than assuming either one will automatically produce the other. A further sharp implication follows from the burnout finding: since it is the amount someone is actively withholding, not merely how much they voice, that predicts burnout, organisations that only track and reward voice frequency may be missing the more consequential half of the picture, and may need to actively seek out what people are withholding rather than inferring wellbeing from how much people are already saying.