Compassion at Work — Research and Evidence

This page gathers what the workplace research actually establishes about compassion: what the construct is, whether it can be trained, where empathy turns counterproductive in leadership, and how strongly care, organizational support, and supportive leadership relate to commitment, retention, and psychological safety. It is a definitional and evidence page — the dollar value, ROI, and cost figures sit on the stakes page, and the manager levers sit on the practice page.

The findings below are deliberately reported to their evidence. Some are well-validated and replicated across hundreds of studies; some are robust correlations that should not be read as proven causation; one — trainability — rests mostly on lab and meditation studies and is best treated as directional. Each section states its own confidence level rather than flattening them into a single claim.

What is compassion at work — and how is it different from empathy and plain kindness?

Compassion at work is an interpersonal process, not a feeling: noticing another person’s pain, feeling an emotional reaction to it, and responding to ease it. Empathy is feeling with someone — sharing their affective state; compassion adds the action tendency. Plain kindness is a general prosocial disposition not necessarily triggered by suffering.

The dominant organizational definition treats compassion as a process. Kanov and colleagues (2004), at the University of Michigan’s “CompassionLab,” specify three subprocesses: noticing another’s pain, feeling an emotional reaction to it, and responding to ease it. This framework has become the field’s standard and is heavily cited, giving the construct strong consensus standing.

The distinction that matters: empathy is feeling with someone — sharing their affective state; compassion adds the action tendency — a motivation to help and the behaviour that follows. Goetz, Keltner and Simon-Thomas (2010), a multidisciplinary review in Psychological Bulletin, conclude compassion is a distinct emotion with its own appraisal pattern (oriented to undeserved suffering), its own signalling (caregiving touch, posture, voice), and its own physiology (social approach rather than withdrawal). Sympathy and pity sit in the same emotion family but differ on finer appraisals. Plain kindness, by contrast, is a general prosocial disposition not necessarily triggered by suffering.

The construct itself is verified; what is weaker is any claim about how much compassion shifts hard business outcomes (covered in the sections below). The practical payload for a manager is that compassion is a behavioural sequence that can actually be performed — notice, feel, act — not a fixed personality trait someone either has or lacks. For a small or mid-sized employer this reframes “being compassionate” from a vague virtue into three concrete managerial moves, which is what makes it trainable and repeatable.

Sources: Kanov, Maitlis, Worline, Dutton, Frost & Lilius, Compassion in Organizational Life, American Behavioral Scientist 47(6), pp. 808–827 (2004); Goetz, Keltner & Simon-Thomas, Compassion: An Evolutionary Analysis and Empirical Review, Psychological Bulletin 136(3) (2010); Worline & Dutton, Awakening Compassion at Work (Berrett-Koehler, 2017).

Confidence: verified for the construct definition.

Can compassion and empathy be trained, or are they fixed traits?

The evidence indicates compassion is a trainable capacity rather than a fixed personality trait. Short-term compassion training measurably shifts brain activation, raises positive affect, and increases prosocial behaviour. The honest limit is that most of this evidence comes from meditation-based training in lab settings, not from workplaces, so the defensible claim is narrow: the capacity is plastic, but a single corporate workshop is not proven to transfer to the shop floor.

A fair question from any employer is whether to train managers in compassion or simply hire for it. The neuroscience says it is trainable. Klimecki and colleagues (2014) showed short-term compassion training shifts brain activation and raises positive affect, and that compassion versus empathy training produce different plasticity patterns — compassion training strengthening reward and affiliation networks. Leiberg, Klimecki and Singer (2011), using a prosocial game, found helping behaviour increased after short-term compassion training but not after a matched memory-training control — early evidence that brief compassion training raises prosocial behaviour toward strangers, on a task unrelated to the training.

The honest weakness, which is why this is treated as directional: most of the evidence comes from meditation-based training in lab settings (largely the Max Planck / University of Zurich line of work), with modest samples, and outcomes measured as affect, brain response, or game behaviour — not manager behaviour or business results. There is no strong evidence that a half-day corporate “compassion workshop” transfers to the shop floor. The defensible claim is narrow: the capacity is plastic, so leaders are not stuck with whatever empathy they were born with.

The practical implication for a small or mid-sized employer is modest but useful: compassion is better treated as a learnable leadership skill — the notice–feel–act sequence can be practised — than as a fixed trait, while keeping expectations realistic about what a single workshop will change.

Sources: Klimecki, Leiberg, Ricard & Singer, Differential pattern of functional brain plasticity after compassion and empathy training, Social Cognitive and Affective Neuroscience 9(6) (2014); Leiberg, Klimecki & Singer, Short-Term Compassion Training Increases Prosocial Behavior in a Newly Developed Prosocial Game, PLoS ONE 6(3):e17798 (2011); Singer & Klimecki, Empathy and compassion, Current Biology 24(18) (2014).

Confidence: directional — the capacity is plastic, but the transfer of brief workshops to workplace behaviour is unproven.

What are the limits and dark side of empathy in leadership?

Empathy — sharing another person’s pain — can exhaust the leader, bias them toward in-group favourites, and distort fair judgment, whereas compassion (care plus reasoned action) is the more sustainable mode. Empathy and compassion recruit different neural systems: sharing pain drives empathic distress, withdrawal, and burnout risk, while compassion engages reward and affiliation circuits and produces positive, approach-oriented affect.

This is the counterweight that keeps the topic from turning sentimental. Singer and Klimecki (2014) summarize training studies showing empathy and compassion recruit different neural systems: sharing another’s pain (empathy) drives empathic distress — negative affect, withdrawal, and burnout risk — whereas compassion engages reward and affiliation circuits and produces positive, approach-oriented affect. The practical claim: empathy exhausts, compassion sustains. These are small-sample fMRI and training studies, so the mechanism is well-evidenced but modest in scope.

Cameron and Payne (2011) demonstrate “compassion collapse”: people felt less compassion for many victims than for one, apparently because they pre-emptively down-regulate feeling when they expect to be overwhelmed. A replication caveat: Hagman and colleagues (2022) contest that mechanism (motivated down-regulation) while noting the collapse effect itself has replicated across 20-plus studies. So the effect should be treated as robust and the mechanism as directional.

Paul Bloom’s Against Empathy (2016) is a trade book making a real scholarly argument, not peer-reviewed research. Bloom argues empathy is a “spotlight” — biased toward the identifiable, the near, and the in-group, and insensitive to scale — and therefore a poor guide for fair decisions; he favours “rational compassion.” For a leader, the three operational dangers are: empathic exhaustion (burnout), in-group favouritism (over-helping the people the leader identifies with), and impaired judgment (one vivid case crowding out the fair, numerate decision).

In a 20-to-200-person firm, where a manager knows everyone personally, empathy’s in-group bias is a live fairness risk — favourites, inconsistent discipline. The remedy the science points to is compassion (care plus action, governed by reason), not more raw feeling.

Sources: Singer & Klimecki, Empathy and compassion, Current Biology 24(18), pp. R875–R878 (2014); Cameron & Payne, Escaping affect: how motivated emotion regulation creates insensitivity to mass suffering, Journal of Personality and Social Psychology 100(1) (2011); Hagman, Tinghög, Dickert, Slovic & Västfjäll, Motivated Down-Regulation of Emotion and Compassion Collapse Revisited, Frontiers in Psychology 13:801150 (2022); Bloom, Against Empathy: The Case for Rational Compassion (Ecco/HarperCollins, 2016).

Confidence: industry-consensus — the empathy/compassion distinction and the collapse effect are well-supported; the underlying mechanism is directional.

Is care a tradeoff with performance, or do compassion and results go together?

Compassion and performance are positively associated, not opposed — but the strongest workplace evidence is correlational and partly drawn from healthcare. There is no credible evidence that compassion costs performance, and consistent evidence that it travels with the attitudes that drive performance, so the defensible verb is “associated with” rather than “causes.”

The “care versus results” framing is largely false in the data — though the evidence is mostly correlational. Lilius and colleagues (2008), the foundational empirical study, found that compassion experienced at work is positively associated with positive emotion and affective commitment — the attitudinal precursors of discretionary effort and retention. Being cross-sectional, it establishes association and mechanism (compassion reshapes how people see co-workers, themselves, and the organization), not causation.

Dutton and Heaphy’s (2003) work on high-quality connections supplies the relational mechanism: brief, respectful, life-giving interactions build the psychological and physiological resourcefulness that supports engagement and learning.

Michael West’s compassionate-leadership program (King’s Fund / NHS) argues that compassion — attending, understanding, empathising, helping — drives innovation, quality, and fewer errors. A clear flag belongs here: this body of work is healthcare-specific and partly practitioner and advocacy in tone; West’s large effect figures are patient-care findings, not findings about managing a small or mid-sized business, and should not be transferred wholesale to a 50-person firm.

The honest bottom line: there is no credible evidence that compassion costs performance, and consistent evidence it travels with the attitudes that drive performance — but a leader should treat “compassion causes higher output by X per cent” as unproven. The defensible verb is “associated with.” That is still a useful result for a small or mid-sized employer: caring for people is not a tax on results, and in practice it moves the commitment and effort small firms depend on — just without promising a specific performance lift.

Sources: Lilius, Worline, Maitlis, Kanov, Dutton & Frost, The contours and consequences of compassion at work, Journal of Organizational Behavior 29(2), pp. 193–218 (2008); Dutton & Heaphy, The Power of High-Quality Connections, in Positive Organizational Scholarship (Berrett-Koehler, 2003); West & Chowla, Compassionate leadership for compassionate health care (King’s Fund / Routledge, 2017).

Confidence: industry-consensus as correlation; weaker as proven causation.

Does feeling supported by the organization improve commitment, effort, and retention?

Yes — perceived organizational support is one of the best-evidenced constructs in the field. It is strongly correlated with affective commitment, moderately with performance, and strongly and negatively with intention to quit, and supervisor support is its single strongest lever.

This is the strongest-evidenced link in the topic. Organizational support theory (Eisenberger and colleagues, 1986) holds that employees form a generalized perception of whether the organization values their contributions and cares about their well-being — perceived organizational support (POS) — and reciprocate it. Two large meta-analyses put numbers on it.

Rhoades and Eisenberger (2002) report corrected correlations of POS with affective commitment (.73), job satisfaction (.62), overall performance (.20), and withdrawal (−.34) — with the link to turnover intentions the strongest at −.51. Kurtessis and colleagues (2017), across 558 studies, closely replicate this: affective commitment .69, in-role performance .19, turnover intentions −.50, and — critically — supervisor support → POS at .60: supervisors, seen as agents of the organization, are the single strongest driver of whether people feel the organization cares.

These magnitudes should be read honestly: the attitudinal effects (commitment, satisfaction, intention to stay) are strong; the performance effect is small-to-moderate (around .19) — feeling supported shifts loyalty and intent more than raw task output. The turnover-intention link is large; actual turnover is weaker (approximately −.21), because quitting also depends on the job market. These are correlational meta-analyses, but the validated measure and replication across 700-plus studies make this construct verified.

For a small or mid-sized employer, the lever is exactly the one a small firm controls: POS is built mostly by the direct supervisor, not a big-company benefits budget. Perceived support rises with managers who visibly value and care for staff — and the payoff shows up first as lower intention to quit. The dollar value of the turnover this prevents lives in the stakes material.

Sources: Kurtessis, Eisenberger, Ford, Buffardi, Stewart & Adis, Perceived Organizational Support: A Meta-Analytic Evaluation of Organizational Support Theory, Journal of Management 43(6), pp. 1854–1884 (2017); Rhoades & Eisenberger, Perceived organizational support: A review of the literature, Journal of Applied Psychology 87(4) (2002); Eisenberger, Huntington, Hutchison & Sowa, Perceived organizational support, Journal of Applied Psychology 71(3) (1986).

Confidence: verified.

Does supportive or compassionate leadership relate to psychological safety and speaking up?

Supportive, high-quality leader relationships are robustly and positively correlated with psychological safety, which in turn predicts learning, voice, and citizenship — so “correlates with” is the precise and defensible verb. Leadership is one of the strongest antecedent families of psychological safety, but not the only one: autonomy, role clarity, and team design also matter.

The claim that compassionate leadership relates to psychological safety is well-supported as a correlation. Carmeli, Brueller and Dutton (2009) surveyed 212 working adults and found high-quality interpersonal relationships positively associated with psychological safety, which in turn related to more learning behaviour. It is a single correlational, self-report study, so it supports “relates to,” not “causes.”

The meta-analytic backstop is Frazier and colleagues (2017), aggregating 136 samples (around 22,000 individuals). On the antecedent side, “positive leader relations” correlated with psychological safety at about ρ = .44 (transformational leadership approximately .42; inclusive leadership approximately .36) — leadership is one of the strongest antecedent families. On the outcome side, psychological safety correlated with information sharing (approximately .52), learning behaviour (approximately .62), citizenship (approximately .32), commitment (approximately .48), and satisfaction (approximately .53). Voice and speaking-up are confirmed positive in the literature, though the exact pooled value is not reproduced here.

The precise, defensible reading: supportive leadership is one of several antecedents of psychological safety — alongside autonomy, role clarity, and team design — and the relationships are correlational. The lever is real but not the only one: a caring manager raises the odds people speak up, but team structure and role clarity matter too. The psychological-safety construct itself is defined in full on the Belonging, Psychological Safety, and the Manager-Employee Relationship page — cross-linked here, not restated.

For a small or mid-sized employer, the takeaway is that the owner-manager’s everyday warmth is a genuine driver of whether problems surface — while remembering that clear roles and sensible team design do part of the work too.

Sources: Frazier, Fainshmidt, Klinger, Pezeshkan & Vracheva, Psychological Safety: A Meta-Analytic Review and Extension, Personnel Psychology 70, pp. 113–165 (2017); Carmeli, Brueller & Dutton, Learning behaviours in the workplace: the role of high-quality interpersonal relationships and psychological safety, Systems Research and Behavioral Science 26 (2009).

Confidence: industry-consensus as correlation.

---

This page is general information drawn from published research, not advice for a specific situation. The findings are reported to their evidence and confidence levels; effect sizes and the line between correlation and causation should be respected before any claim is made about a particular workplace. Where figures here are described as correlational, they should not be presented as proof that compassion causes a given business result.