Learning to see by learning to draw: A longitudinal analysis of the relationship between representational drawing training and visuospatial skill.

42 first-year art students (37 female), mean age 18.6; three testing points over five months

First-year students at Pratt Institute, an elite US art school, receiving eight hours of drawing instruction per week plus homework

This is the best study in existence on whether art instruction improves art skill, and it is a single-cohort pre-post with NO CONTROL GROUP. That sentence is the finding. Outcome was a ten-minute observational still-life drawing sorted into seven quality categories by ten non-expert and six expert judges blind to identity, with inter-judge reliability 0.98 - a well-constructed researcher-designed measure, but researcher-designed. Grade D under the archive's grid: pre-post without a control group. The authors state the limitation themselves and draw the comparison that matters: their field, they write, is "in stark contrast to the large body of evidence from longitudinal designs suggesting that musical training exerts causal impacts on perceptual and cognitive abilities."

Roughly 160 hours of elite studio instruction over five months produced a linear mixed-effects time coefficient of 0.17 (SE 0.06, t = 2.82, p < .01) on a seven-point judged drawing-quality scale - about a third of a point. Nothing was detectable over the first two months (T1-T2 t(38) = 0.08, p = .94); the change appears between months two and five (T2-T3 t(38) = 2.59, p = .01), and only the T1-T3 contrast survives Bonferroni correction (t(38) = 3.03, p = .004). Without a control arm, practice effects on the repeated drawing task cannot be separated from instruction. The authors' own limitation, verbatim: "A more significant limitation of the current study was that there was no control group in which to measure change in visuospatial performance over time without a drawing intervention."

Unmanaged and structurally unmanageable in this design - the sample is self-selected art students at a selective institution, and the comparison is within-person over time. Arden et al. (2014) put the heritability of children's drawing at 0.29 with a genetic correlation of 0.52 to later g, so the trait being trained here has a substantial heritable component.

No replication and no controlled successor located. The wider drawing-expertise literature (Kozbelt 2001; Chamberlain et al. 2014, 2019; Perdreau and Cavanagh 2014) is cross-sectional expert-versus-novice work that cannot separate training from selection.