Do New Forms of Reading Pay Off? A Meta-Analysis on the Relationship Between Leisure Digital Reading Habits and Text Comprehension

39 effect sizes from 25 reports (one influential outlier removed from 40/26); 469,564 participants; studies published 2000-2022

Predominantly students (97.4% of samples) across primary school, middle school, high school and university, plus two adult samples. Data are largely from large international assessments and survey-linked comprehension tests.

Multilevel (three-level) random-effects meta-analysis with robust variance estimation of the CORRELATION between self-reported leisure digital reading habits (social-communicative reading such as messaging and social media, and linear-informative reading such as news and e-books) and text comprehension. This is a meta-analysis of CROSS-SECTIONAL CORRELATIONS, which is why it is graded D and not C: per METHODOLOGY.md a meta-analysis inherits the design of what it aggregates, and no amount of multilevel machinery converts an association between a self-reported habit and a test score into a causal estimate. Every standard confound is live and unaddressed — children who read more print are different children, in heritable ways, from children who scroll more, and the direction of causation is at least bidirectional (the paper's own literature review cites longitudinal work finding comprehension predicts later print reading habits more strongly than the reverse). Heterogeneity is enormous: I2 = 98.9%, Q(38) = 2591.93, and the 95% PREDICTION interval runs from -.18 to +.29, i.e. the paper cannot rule out a meaningfully negative association in a new population. Publication bias tests were negative (published vs unpublished p = .53), and the five extra-large samples (N > 10,000) did not differ from the rest (p = .15). Independent academic work, University of Valencia; no publisher or device-industry funding declared. DISPLACEMENT WAS NOT MEASURED: the paper argues ABOUT the displacement hypothesis (and concludes its results do not support it, favouring a "shallowing" account instead) but it holds no time-use data, so it cannot say what digital reading replaced.

Leisure digital reading is essentially uncorrelated with reading comprehension: r = .055, 95% CI [.003, .107], p = .03, across 39 effect sizes and 469,564 participants — statistically significant only because the sample is enormous, and an order of magnitude smaller than the r = .36-.41 that prior meta-analyses report for leisure PRINT reading. The pattern by school stage is the part that bites: the association is NEGATIVE in primary and middle school (middle school r = -.025) and turns positive only in high school (r = +.085) and university, with educational stage a significant moderator (p = .047, R2 = .57). Type of digital reading did not matter significantly (social-communicative r = .03 vs linear-informative r = .12, p = .18), which undercuts the hopeful reading that the problem is only social media rather than screen reading generally. Read against Delgado et al. (2018), the two say the same thing from opposite methods: the medium costs comprehension when you experiment on it, and it buys almost nothing when you correlate habits with outcomes. Read on its own, this is correlational evidence and cannot bear a causal claim in either direction.

High. Reading habits and reading comprehension are both substantially heritable and share family environment; the same children who read print books have parents who supply both books and alleles. No study in the corpus is within-family, and none is randomized.

We have not checked the literature for an independent re-analysis of this meta-analysis, so this is recorded as `unclear` rather than `unreplicated`, which would assert a fact about the literature we cannot source. Two things are worth separating. The COMPARISON it draws on — that leisure PRINT reading correlates with comprehension at roughly r = .36-.41 — comes from prior meta-analyses (Mol & Bus 2011 and successors) and is well established. The paper's OWN estimate (r = .055 for digital leisure reading) is a single synthesis, and its most striking sub-result, the negative association at primary and middle school, rests on a moderator contrast that reached p = .047 with only two significant pairwise comparisons. Delgado et al. (2018) is a companion rather than a replication: it tests the medium experimentally, this tests reading HABITS observationally, and neither can stand in for the other.