n/a · idea · research

Dual coding, multimedia learning and the learning-styles myth

The evidence that words plus relevant pictures beat words alone, moderately, and that 'visual learners' is a myth.

inspireevidence: strongby Paivio; Mayer; Sweller; Pashler et al.; Cromley and Chendoi.org/10.1016/j.edurev.2025.100730 ↗

Visual against text learning: dual coding, multimedia learning, and the learning-styles myth

  • Maker: a research field. Allan Paivio (dual coding theory, 1971); Richard Mayer (cognitive theory of multimedia learning, 1990s onward); John Sweller (cognitive load); critiques of learning styles by Pashler et al. (2008).
  • URL: https://doi.org/10.1016/j.edurev.2025.100730 (the most recent meta-analysis we found)
  • Status (10 October 2026): an active field with a recent reckoning. Cromley and Chen’s meta-analysis of Mayer’s own studies (Educational Research Review, November 2025) and Noetel et al.’s meta-meta-analysis (Review of Educational Research 92(3), 2022) are the current summaries.

What it is

The evidence behind the claim that people learn better by seeing. In short:

  • Dual coding (Paivio): verbal and visual information are processed by two partly separate systems; material encoded in both is remembered better.
  • Multimedia learning (Mayer): “people learn better from words and pictures than from words alone,” with a set of design principles: coherence (remove extraneous material), signalling, spatial and temporal contiguity, modality (narration plus visuals rather than on-screen text), segmenting, pre-training, personalisation.
  • Learning styles (“I’m a visual learner”): matching instruction to a preferred modality. This is not supported. Pashler et al. (2008) found almost no studies meeting the design needed to show it, and those that did mostly didn’t support it. Belief in it stays widespread among teachers.

The problem it’s solving

What actually helps people understand, and why.

Its path / bet

Controlled experiments, mostly short lab lessons with transfer and retention tests.

How it works (concretely): what the numbers say

  • Cromley and Chen (2025), across Mayer’s multimedia research from 1990 to 2022: an overall effect of g = 0.37 over 181 studies. The largest reliable effects came from removing seductive details (interesting but irrelevant pictures and facts), the modality principle, personalisation, the multimedia principle itself, sentence-level coherence and self-explanation. Effects have declined slightly per year of publication, and are smaller or not significant for elementary and graduate students. Games and simulations averaged about g = 0.41, larger for inference and transfer than for recall (as reported in Carl Hendrick’s summary; we read the summary, not the full paper).
  • Fiorella and Mayer’s 2022 review of the modality principle reports a median d = 0.65 across 76 studies. Mayer’s 2017 review gives d = 1.67 for the multimedia principle but from only five comparisons; treat that as unstable.

Strengths

  • Decades of experiments; a moderate, consistent benefit of relevant visuals with words.
  • Clear about what hurts: decoration. Irrelevant visuals lower learning.

Weaknesses / limits

  • Mostly short lab studies with university students; effects shrink in newer, more varied work.
  • The field studied authored, carefully designed material. Whether generated visuals of uneven quality get the same benefit is untested.
  • “Visual” in this research means a relevant picture with words, not “interactive app.”

Relation to fictty

Inspire, and a check on our own story. The evidence supports “words plus relevant visuals beat words alone,” modestly. It does not support “people are visual learners,” and it warns that decorative visuals hurt. Generated UIs are at high risk of seductive detail: they’re impressive and often irrelevant.

Could fictty adopt it instead of building?

Not applicable.

What fictty should take from it

  • Say it carefully in the blog post. Claim the multimedia principle (a moderate, replicated effect), not learning styles. Don’t say “humans are visual.”
  • Coherence is a design rule for generated screens. Fewer, relevant elements; no decorative pixels. fictty’s guidance (“components first, animate the data not the pictures”) points the right way; add “nothing that isn’t part of the explanation.”
  • Contiguity is a layout rule. Put the explanation next to the thing it explains: callouts on the node, not a paragraph below. fictty’s layers (callouts, spotlights, arrows) are exactly this.
  • Interactive simulations show the larger effects on transfer. That’s the case for screens you can manipulate, not just look at.

Sources