Results.

<p><b>(A)</b> Mean azimuthal errors as a function of trial number for the three magnification conditions (<i>m</i> = 2, <i>m</i> = 1, <i>m</i> = 0.5), together with a model (<a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0117901#pone.0117901.e005" target="_blank">eq. 3</a>) with the mean fit parameters of the 13 subjects. Gray background indicates the blocks without visual feedback. <b>(B)</b> Model comparison. Same data as in A, with a one-process model (<i>θ</i>_<i>i</i>+1 = <i>A*θ</i>_<i>i</i> + <i>B * e</i>_<i>i</i>) with mean fit parameters. <b>(C)</b> Comparison of the Akaike information Criterion for the two-process model (AIC₂) and one-process model (AIC₁). For all subjects, the AIC was lower for the two-process model, indicating that it was a better description of the data. <b>(D)</b> Fit learning fractions (<i>B</i>₁, dark grey bars; <i>B</i>₂, light grey bars) with 95% confidence intervals. <b>(E)</b> Fit retention fractions (<i>A</i>₁, dark grey bars; <i>A</i>₂, light grey bars) with 95% confidence intervals.