Lead
After two children clash, parents tend to reach for the same words: "Say sorry."
The child says it, reluctantly. The moment passes. But how much did the child actually grasp — what happened to the other child, why the act was wrong, how the other person experienced it? Does requiring "sorry" build empathy? Or does the word drift off as ritual, unanchored to anything felt?
This article draws on empirical research to map how empathy develops, and to think through what "sorry" actually means at different ages. The short version: empathy moves sharply in the second year of life, but the social understanding that "sorry" requires tends to come somewhat later — and the two should not be conflated.
When Empathy Begins, and in What Order
The most widely cited framework for empathy development is Martin Hoffman's four-stage model [1]. In the first year, Hoffman describes global empathy: the infant does not yet distinguish its own distress from another's, so another baby's cry triggers its own crying [1]. Around the second year, egocentric empathy emerges — the child notices another person's distress and responds, but the response is organized around what would comfort the child itself (offering their own stuffed animal, for instance) [1]. From the third year onward, empathy for another's feelings develops, enabling comfort that is genuinely calibrated to the other person's perspective [1].
The observational evidence for this sequence comes largely from Carolyn Zahn-Waxler and her colleagues' longitudinal research [2]. Watching children from 13 to 24 months, they documented a substantial increase in empathic responding to others' distress across that window. Reparative behavior — actions toward someone the child has inadvertently hurt — also increased over the same period, with virtually all children showing some form of helping behavior by age two [2]. Crucially, the study showed that these developments coincide with the parallel emergence of self-recognition (the ability to recognize one's own reflection in a mirror) [2]. To comfort another, you first need to distinguish self from other — the developmental sequence has that logic built in.
At the behavioral level, the most elegant demonstration comes from Felix Warneken and Michael Tomasello, who reported in Science in 2006 that 18-month-old infants spontaneously pick up objects an adult has dropped — without prompting, without reward — while declining to help when the drop was clearly intentional [3]. Prosocial, other-oriented behavior appears very early, and it appears without being taught.
Understanding Another Mind Comes Later
Here a careful distinction is needed. Responding to another's distress and understanding that another person holds different beliefs from your own are two different capacities. The second is what developmental psychologists call Theory of Mind: the cognitive ability to attribute mental states — beliefs, desires, intentions — to oneself and others, and to understand that others' mental states may differ from one's own, typically measured with false-belief tasks: experiments testing whether a child understands that another person can hold a belief that the child knows to be incorrect.
Henry Wellman, David Cross, and Julianne Watson's meta-analysis of 178 studies, published in Child Development in 2001, established the developmental timeline for false-belief understanding with unusual precision [4]. Performance on false-belief tasks is at or below chance in three-year-olds, rises reliably above chance between ages four and five, and this pattern holds across cultures and task formats [4]. The capacity to understand that another person can hold a wrong belief — one that differs from reality and from the child's own knowledge — becomes solidly established around age four.
What does this imply? A two- or three-year-old can notice that another child looks hurt or sad. But reasoning about how their own action appeared from the other child's perspective, or inferring the other child's intention, is still beyond what most children this age can reliably do [1,2,4]. Ask a toddler "how do you think she felt when you hit her?" and silence is not evidence of callousness — the neural scaffolding for that inference is still being assembled.
What "Sorry" Is Actually Doing
Smith, Chen, and Harris investigated when children understand the interpersonal function of apology — its role in soothing the feelings of the person who was hurt [5]. Their research found that placing apology within that relational frame, and recognizing it as a tool for repairing the victim's emotional state, tends to come at four and older. Below that age, "sorry" is more likely functioning as a situational script — a phrase that ends the episode — than as a genuine act of repair [5].
What develops earlier than the verbal apology is the behavioral equivalent. Amrisha Vaish and colleagues ran experiments with two-year-olds in which the child's action inadvertently harmed another person [6]. Children showed signals consistent with something like guilt — downward gaze, subdued manner — followed by helping behavior. Behavioral repair precedes verbal apology in the developmental sequence. This finding aligns squarely with both Hoffman's stages and Zahn-Waxler's observational work [1,2].
The implication is uncomfortable for a common parenting reflex: requiring "sorry" from a two-year-old is unlikely to cultivate empathy. Nancy Eisenberg's review of the prosocial development literature is relevant here [7]. Her conclusion is that parental emotion coaching — putting feelings into words, narrating what is happening for both parties — has a meaningful association with children's prosocial development. Describing what happened is more developmentally useful scaffolding than demanding a word.
What This Looks Like in Practice
Research findings translate into a different set of moves for the moment after a two- or three-year-old clash:
- Before "say sorry," narrate: "Looks like she got hurt." "He seems startled." Label what's observable.
- If the child shows a repair impulse — returning a toy, moving closer to look — name it: "You're checking on her. That's kind."
- Don't read the absence of "sorry" as the absence of empathy. The capacity to understand when that word is warranted — and what it means to the person receiving it — is still developing [4,5].
- Speak to the other child too: "You were sad, weren't you?" This makes both emotional states visible and separate, which is part of how children learn to hold two minds at once.
These moves are not designed to abolish "sorry" from a child's life. The goal is to build the soil in which the social convention of apology can later take root and mean something. Teaching the word and building the understanding it requires are two different projects, running on different timelines.
It is also worth noting what Eisenberg's framework implies about tone [7]. Emotion coaching is not a therapy technique requiring special training; it is simply narrating the emotional facts of a situation as they unfold. "She's crying because her block fell" communicates more useful information to a two-year-old than "say sorry" does, because it grounds the interaction in observable cause and effect rather than social convention. That kind of running commentary — matter-of-fact, without blame — is what the research associates with prosocial development over time. There is no need to make it elaborate or to do it every time. The habit matters more than the performance.
A Note on Recording
When "hit" and "made cry" are all that get logged, the whole day tips toward an impression of aggression. Keeping a record in an app like Memori — noting not just the clash but the repair gestures and the moments of comfort that followed — makes the early signs of empathy visible. Development is continuous. Reading it as a line rather than a point gives a truer picture of what is actually growing.
Summary
Empathy develops substantially around age two, and by age three virtually all children show some form of prosocial helping behavior [2]. The capacity to reason from another's perspective — Theory of Mind — stabilizes around age four [4], and the social-function understanding of "sorry" tends to arrive at roughly the same time [5].
At two and three, teaching "sorry" as a social convention is fine. Using it as a measure of empathy is not. Empathy is already in motion, slightly earlier, in a different register entirely.
References
- Hoffman ML. Empathy and Moral Development: Implications for Caring and Justice. Cambridge University Press; 2000. doi:10.1017/CBO9780511805851.
- Zahn-Waxler C, Radke-Yarrow M, Wagner E, Chapman M. Development of concern for others. Dev Psychol. 1992;28(1):126–136. doi:10.1037/0012-1649.28.1.126.
- Warneken F, Tomasello M. Altruistic helping in human infants and young chimpanzees. Science. 2006;311(5765):1301–1303. doi:10.1126/science.1121448. PMID: 16513986.
- Wellman HM, Cross D, Watson J. Meta-analysis of theory-of-mind development: the truth about false belief. Child Dev. 2001;72(3):655–684. doi:10.1111/1467-8624.00304. PMID: 11405571.
- Smith CE, Chen D, Harris PL. When the happy victimizer says sorry: children's understanding of apology and emotion. Br J Dev Psychol. 2010;28(Pt 4):727–746. doi:10.1348/026151009X475343.
- Vaish A, Carpenter M, Tomasello M. Young children's responses to guilt displays. Dev Psychol. 2011;47(5):1248–1262. doi:10.1037/a0024462.
- Eisenberg N, Spinrad TL, Sadovsky A. Empathy-related responding in children. In: Killen M, Smetana JG, editors. Handbook of Moral Development. Mahwah, NJ: Erlbaum; 2006. p. 517–549.