Reading literary fiction or taking part in prosocial reading was associated with a small positive link to empathy-related outcomes, according to a systematic review that combined 21 primary studies. In the quantitative analysis, 12 studies supplied 14 effect sizes. The pooled estimate was Hedges’ g = 0.23, with a 95% confidence interval from 0.11 to 0.35. Hedges’ g is a standardized measure that puts results from different tests on the same scale. The result points to a modest association, not proof that reading caused the difference. Because the review combined experimental, quasi-experimental, correlational and neurocognitive research, the evidence does not carry one uniform level of causal certainty.
Where the signal appeared
The cognitive-empathy and theory-of-mind subgroup showed a small positive pooled effect. Across eight effect sizes, the estimate was Hedges’ g = 0.27, with a 95% confidence interval of 0.10 to 0.44. The review also reported substantial variation between those results, with I-squared at 84.2%. In practical terms, the studies did not point to one consistent size of association, even though the combined estimate was positive.
Measures of affective empathy and prosocial outcomes also showed a small positive pooled association, with Hedges’ g = 0.19 across six effect sizes. The 95% confidence interval ran from 0.06 to 0.32, and I-squared was 60.3%. But direct comparisons in this group were often small or inconclusive, so the result should not be read as evidence of a large or uniform change in behaviour.
No clear winner
The review did not find a statistically significant gap between the two outcome groups. The pooled estimates were 0.27 for cognitive and theory-of-mind outcomes and 0.19 for affective and prosocial outcomes, with the between-group test giving p = 0.34. Nor did literary fiction clearly outperform prosocial reading. Their pooled estimates were 0.24 and 0.20, respectively, and that comparison gave p = 0.54. The authors describe both tests as exploratory, so the numbers should not be read as proof that one kind of reading is better.
A mixed pattern
The broader pattern was mixed. Of the 21 studies in the qualitative synthesis, 10, or 47.6%, were categorized as positive. Six studies, or 28.6%, reported mediated or conditional findings, while five, or 23.8%, were null or weak. Three studies, or 14.3%, were classified as neurocognitive. These categories could overlap, so they are not four mutually exclusive piles of evidence. Positive findings were the largest group, but they did not account for the whole literature reviewed.
The review’s account of affective and prosocial outcomes points to conditions around reading as much as to the label placed on a text. Responses were described as depending on narrative transportation, or becoming absorbed in a story, identification with characters, explicit instructions to take another person’s perspective, or characteristics of the reader, rather than arising uniformly from text type alone. That pattern helps explain why the pooled result was positive while the individual study findings varied.
Why confidence should stay measured
The review searched five databases from January 2000 to early 2025 and checked reference lists and forward citations for additional studies. It included experimental, quasi-experimental, correlational and neurocognitive designs, bringing together evidence on literary and prosocial reading. Those different designs sit inside an overall result with substantial heterogeneity, meaning the size of the reported association varied considerably across the evidence.
The category-level risk assessment rated preregistered studies low across domains, randomized experiments moderate overall, correlational studies high overall and neuroimaging studies moderate overall. The review itself was not prospectively registered, although it said eligibility, selection, outcome classification and extraction procedures had been defined before formal data extraction. Those details matter because the pooled estimate is built from evidence with different risk profiles.
Tests for publication bias did not show statistically significant asymmetry in the 14 effect sizes examined. Egger’s test reported an intercept of 0.152 with p = 0.14, while Begg’s rank-correlation test gave tau = 0.18 with p = 0.27. The review cautioned that the small number of effects limits the power of those tests, so the result cannot rule out selective publication.
The cautious takeaway
Taken together, the review describes reading as linked to modest, context-sensitive differences in empathy-related outcomes, particularly in cognitive empathy and theory of mind. It does not support treating literary fiction as an automatic route to empathy: the overall analysis combined literary and prosocial reading, found no significant advantage for either exposure, and described affective and prosocial responses as depending on engagement and reader factors. The practical message is restrained. Reading may matter, but the evidence does not support expecting a large or uniform shift from text type alone.
Paper data and sources
Original title: Empathy and perspective-taking in literary fiction: a systematic review and meta-analysis of psychological evidence and theoretical insights
Authors: Chang L
Journal/Repository: Frontiers in psychology
Status: Peer-reviewed
First online: 2026-10-10
DOI: Not available
Original paper · Full text